Travis Oliphant: NumPy, SciPy, Anaconda, Python & Scientific Programming

Lex Fridman (01:20.920)

BASIC. BASIC, yeah, on an Atari 800,

Travis Oliphant (01:23.320)

Atari 400, I think, or maybe it was an Atari 800.

Lex Fridman (01:26.840)

It was a part of a class,

Lex Fridman (01:28.300)

and we just were just BASIC loops to print things out.

Lex Fridman (01:32.560)

Did you use go to statements?

Travis Oliphant (01:34.920)

Yes, yes, we used go to statements.

Lex Fridman (01:38.000)

I remember in the early days,

Travis Oliphant (01:39.560)

that's when I first realized

Lex Fridman (01:41.160)

there's like principles to programming,

Travis Oliphant (01:43.320)

when I was told that don't use go to statements.

Lex Fridman (01:45.720)

Those are bad software engineering principles,

Travis Oliphant (01:48.080)

like it goes against what great, beautiful code is.

Lex Fridman (01:52.040)

I was like, oh, okay, there's rules to this game.

Travis Oliphant (01:54.800)

I didn't see that until high school

Lex Fridman (01:56.240)

when I took an AP computer science course.

Travis Oliphant (01:58.360)

I did a lot of other kinds of just programming in TI,

Lex Fridman (02:02.200)

but finally, when I took an AP computer science course

Travis Oliphant (02:04.160)

in Pascal.

Lex Fridman (02:05.720)

Wow.

Travis Oliphant (02:06.560)

That's, yeah, it was Pascal.

Lex Fridman (02:07.440)

That's when I, oh, there are these principles.

Lex Fridman (02:09.760)

Not C or C++?

Lex Fridman (02:11.320)

No, I didn't take C until the next year in college.

Travis Oliphant (02:14.660)

I had a course in C, but I haven't done much in Pascal,

Lex Fridman (02:18.100)

just that AP computer science course.

Travis Oliphant (02:20.160)

Now, sorry for the romanticized question,

Lex Fridman (02:23.480)

but when did you first fall in love with programming?

Travis Oliphant (02:26.720)

Oh, man, good question.

Lex Fridman (02:27.880)

I think actually when I was 10,

Travis Oliphant (02:30.280)

my dad got us a TI Timex Sinclair,

Lex Fridman (02:33.460)

and he was excited about the spreadsheet capability,

Lex Fridman (02:37.200)

and then, but I made him get the basic,

Lex Fridman (02:39.560)

the add ons we could actually program in basic,

Lex Fridman (02:41.840)

and just being able to write instructions

Lex Fridman (02:44.520)

and have the computer do something.

Travis Oliphant (02:45.960)

Then we got a TI 994A when I was about 12,

Lex Fridman (02:50.080)

and I would just, it had sprites and graphics and music.

Travis Oliphant (02:52.960)

You could actually program it to do music.

Lex Fridman (02:55.320)

That's when I really sort of fell in love with programming.

Lex Fridman (02:58.600)

So this is a full, like a real computer

Lex Fridman (03:01.060)

with like, with memory and storage,

Travis Oliphant (03:04.120)

processors and whatnot,

Lex Fridman (03:05.240)

because you say TI. Yeah, the Timex Sinclair

Travis Oliphant (03:07.360)

was one of the very first, it was a cheap, cheap,

Lex Fridman (03:09.680)

like, I think it was, well, it was still expensive,

Lex Fridman (03:12.760)

but it was 2K of memory.

Lex Fridman (03:14.440)

We got the 16K add on pack,

Lex Fridman (03:16.760)

but yeah, it had memory, and you could program it.

Lex Fridman (03:19.000)

You had the, in order to store your programs,

Travis Oliphant (03:20.920)

you had to attach a tape drive.

Lex Fridman (03:22.880)

Remember that old, the sound that would play

Travis Oliphant (03:24.400)

when you converted the modems would convert digital bits

Lex Fridman (03:29.440)

to audio files set on a tape drive.

Travis Oliphant (03:31.920)

Still remember that sound, but that was the storage.

Lex Fridman (03:34.760)

And what was the programming language, do you remember?

Travis Oliphant (03:36.480)

It was basic. It was basic.

Lex Fridman (03:37.320)

And then they had a VisiCalc,

Lex Fridman (03:38.980)

and so a little bit of spreadsheet programming

Lex Fridman (03:40.600)

in VisiCalc, but mostly just some basic.

Lex Fridman (03:42.760)

Do you remember what kind of things drew you to programming?

Lex Fridman (03:46.340)

Was it working with data, was it video games?

Lex Fridman (03:50.360)

Games, math, mathy stuff?

Lex Fridman (03:52.600)

Yeah, I've always loved math,

Lex Fridman (03:54.800)

and a lot of people think they don't like math

Lex Fridman (03:58.080)

because I think when they're exposed to it early,

Travis Oliphant (04:00.440)

it's about memory.

Lex Fridman (04:02.080)

When you're exposed to math early,

Travis Oliphant (04:03.260)

you have a good short term memory,

Lex Fridman (04:04.280)

can remember his timetables.

Lex Fridman (04:05.920)

And I do have a reasonably, I mean, not perfect,

Lex Fridman (04:08.600)

but a reasonably long little short term memory buffer.

Lex Fridman (04:12.480)

And so I did great at timetables.

Lex Fridman (04:14.320)

I said, oh, I'm good at math.

Lex Fridman (04:15.840)

But I started to really like math,

Lex Fridman (04:17.360)

just the problem solving aspect.

Lex Fridman (04:20.320)

And so computing was problem solving applied.

Lex Fridman (04:25.040)

And so that's always kind of been the draw,

Travis Oliphant (04:28.280)

kind of coupled with the mathematics.

Lex Fridman (04:30.480)

Did you ever see the computer as like an extension

Lex Fridman (04:33.920)

of your mind, like something able to achieve?

Lex Fridman (04:36.520)

Not till later.

Travis Oliphant (04:37.760)

Okay.

Lex Fridman (04:38.600)

Yeah, not then.

Travis Oliphant (04:39.440)

It's just like a little set of puzzles

Lex Fridman (04:40.880)

that you can play with and you can play with math puzzles.

Travis Oliphant (04:43.520)

Yeah, it was too rudimentary early on.

Lex Fridman (04:46.120)

Like it was sort of, yeah, it was a lot of work

Travis Oliphant (04:49.160)

to actually take a thought you'd have

Lex Fridman (04:51.440)

and actually get it implemented.

Lex Fridman (04:53.120)

And that's still work, but it's getting easier.

Lex Fridman (04:56.020)

And so yeah, I would say that's definitely

Travis Oliphant (04:58.240)

what's attracting me to Python

Lex Fridman (04:59.560)

is that that was more real, right?

Travis Oliphant (05:02.140)

I could think in Python.

Lex Fridman (05:04.840)

Speaking of foreign language,

Travis Oliphant (05:05.800)

I only speak another language fluently besides English,

Lex Fridman (05:08.400)

which is Spanish.

Lex Fridman (05:09.220)

And I remember the day when I would dream in Spanish

Lex Fridman (05:11.720)

and you start to think in that language.

Lex Fridman (05:13.440)

And then you actually, I do definitely believe

Lex Fridman (05:15.340)

that language limits or expands your thinking.

Travis Oliphant (05:19.640)

There's some languages that actually lead you

Lex Fridman (05:21.600)

to certain thought processes.

Travis Oliphant (05:23.860)

Yeah, like, so I speak Russian fluently

Lex Fridman (05:27.280)

and that's certainly a language that leads you

Travis Oliphant (05:30.960)

down certain thought processes.

Lex Fridman (05:33.240)

Well, yeah, I mean, there's a history

Travis Oliphant (05:36.220)

of the two world wars of millions of people starving

Lex Fridman (05:41.220)

to death or near to death throughout its history

Travis Oliphant (05:44.180)

of suffering, of injustice, like this promise sold

Lex Fridman (05:48.020)

to the people and then the carpet

Travis Oliphant (05:50.900)

or whatever is swept from under them.

Lex Fridman (05:53.340)

And it's like broken promises.

Lex Fridman (05:54.660)

And all of that pain and melancholy is in the language,

Lex Fridman (05:58.100)

the sad songs, the sad hopeful songs,

Travis Oliphant (06:01.700)

the over romanticized, like, I love you, I hate you,

Lex Fridman (06:05.260)

the sort of the swings between all the various spectrums

Travis Oliphant (06:09.980)

of emotion, so that's all within the language.

Lex Fridman (06:13.740)

The way it's twisted, there's a strong culture

Travis Oliphant (06:18.020)

of rhyming poetry, so like the bards,

Lex Fridman (06:20.380)

like the sync, there's a musicality to the language too.

Lex Fridman (06:24.740)

Did Dostoevsky write in Russian?

Lex Fridman (06:27.380)

Yeah, so like Dostoevsky, Tostoy, all the,

Travis Oliphant (06:32.100)

all the.

Lex Fridman (06:32.940)

The ones that I know about, which are translated

Lex Fridman (06:34.660)

and I'm curious how the translations.

Lex Fridman (06:36.340)

So Dostoevsky did not use the musicality

Travis Oliphant (06:40.860)

of the language too much.

Lex Fridman (06:42.180)

So it actually translates pretty well

Travis Oliphant (06:44.180)

because it's so philosophically dense

Lex Fridman (06:46.540)

that the story does a lot of the work,

Lex Fridman (06:48.460)

but there's a bunch of things that are untranslatable.

Lex Fridman (06:51.140)

Certainly the poetry is not translatable.

Travis Oliphant (06:53.580)

I actually have a few conversations coming up offline

Lex Fridman (06:57.940)

and also in this podcast with people

Travis Oliphant (06:59.980)

who've translated Dostoevsky.

Lex Fridman (07:01.940)

And that's for people who worked, who work in this field,

Travis Oliphant (07:06.340)

know how difficult that is.

Lex Fridman (07:07.340)

Sometimes you can spend months thinking

Lex Fridman (07:10.660)

about a single sentence, right?

Lex Fridman (07:12.340)

In context, like, cause there's just the magic

Travis Oliphant (07:15.220)

captured by that sentence and how do you translate

Lex Fridman (07:17.860)

just in the right way?

Travis Oliphant (07:18.940)

Because those words can be really powerful.

Lex Fridman (07:22.380)

There's a famous line,

Travis Oliphant (07:24.300)

beauty will save the world from Dostoevsky.

Lex Fridman (07:27.140)

You know, there's so many ways to translate that.

Lex Fridman (07:29.500)

And you're right, the language gives you the tools

Lex Fridman (07:32.700)

with which to tell the story,

Lex Fridman (07:34.140)

but it also leads your mind down certain trajectories

Lex Fridman (07:37.260)

and paths to where over time,

Travis Oliphant (07:39.660)

as you think in that language,

Lex Fridman (07:41.140)

you become a different human being.

Travis Oliphant (07:42.740)

Yes. Yeah.

Lex Fridman (07:43.740)

Yeah, that's a fascinating reality, I think.

Travis Oliphant (07:45.860)

I know people have explored that,

Lex Fridman (07:47.020)

but it's just rediscovered.

Travis Oliphant (07:49.740)

Well, we don't, we live in our own like little pockets.

Lex Fridman (07:52.340)

Like this is the sad thing is I feel like unfortunately,

Travis Oliphant (07:56.860)

given time and given getting older,

Lex Fridman (07:59.140)

I'll never know China, the Chinese world,

Travis Oliphant (08:03.620)

because I don't truly know the language.

Lex Fridman (08:05.780)

Same with Japanese, I don't truly know Japanese

Lex Fridman (08:08.300)

and Portuguese and Brazil,

Lex Fridman (08:10.340)

that whole South American continent.

Travis Oliphant (08:12.060)

Like, yeah, I'll go to Brazil and Argentina,

Lex Fridman (08:14.460)

but will I truly understand the people

Lex Fridman (08:17.100)

if I don't understand the language?

Lex Fridman (08:18.500)

It's sad because I wonder how much,

Lex Fridman (08:23.500)

how many geniuses were missing

Lex Fridman (08:25.220)

because so much of the scientific world,

Lex Fridman (08:28.540)

so much of the technical world is in English,

Lex Fridman (08:31.460)

and so much of it might be lost

Travis Oliphant (08:33.140)

because it's just we don't have the common language.

Lex Fridman (08:36.100)

I completely agree.

Travis Oliphant (08:36.940)

I'm very much in that vein of there's a lot of genius

Lex Fridman (08:40.620)

out there that we miss,

Lex Fridman (08:41.780)

and it's sort of fortunate when it bubbles up

Lex Fridman (08:45.060)

into something that we can understand or process,

Travis Oliphant (08:48.700)

there's a lot we miss.

Lex Fridman (08:50.420)

So I tend to lean towards really loving democratization

Travis Oliphant (08:54.060)

or things that empower people

Lex Fridman (08:55.420)

or very resistant sort of authoritarian structures.

Travis Oliphant (09:00.140)

Fundamentally for that reason,

Lex Fridman (09:01.900)

well, several reasons, but it just hurts us.

Travis Oliphant (09:04.740)

We're soft.

Lex Fridman (09:06.420)

So speaking of languages that empower you,

Lex Fridman (09:09.020)

so Python was the first language for me

Lex Fridman (09:11.820)

that I really enjoyed thinking in, as you said.

Travis Oliphant (09:16.780)

Sounds like you shared my experience too.

Lex Fridman (09:18.500)

So when did you first,

Lex Fridman (09:19.620)

do you remember when you first kind of connected with Python,

Lex Fridman (09:21.900)

maybe even fell in love with Python?

Travis Oliphant (09:23.740)

It's a good question.

Lex Fridman (09:24.580)

It was a process.

Travis Oliphant (09:25.500)

It took about a year.

Lex Fridman (09:26.540)

I first encountered Python in 1997.

Travis Oliphant (09:29.500)

I was a graduate student studying biomedical engineering

Lex Fridman (09:31.740)

at the Mayo Clinic.

Lex Fridman (09:32.980)

And I had previously,

Lex Fridman (09:34.700)

I'd been involved in taking information from satellites.

Travis Oliphant (09:39.340)

I was an electrical engineering student

Lex Fridman (09:41.340)

used to taking information

Lex Fridman (09:42.660)

and trying to get something out of it,

Lex Fridman (09:44.060)

doing some data processing, getting information out of it.

Lex Fridman (09:46.140)

And I'd done that in MATLAB.

Lex Fridman (09:47.660)

I'd done that in Perl.

Travis Oliphant (09:49.140)

I'd done that in scripting on a VMS.

Lex Fridman (09:52.540)

There's actually a VAX VMS system,

Travis Oliphant (09:54.260)

they had their own little scripting tools around Fortran.

Lex Fridman (09:57.980)

Done a lot of that.

Lex Fridman (09:58.820)

And then as a graduate student,

Lex Fridman (10:00.820)

I was looking for something and encountered Python.

Lex Fridman (10:04.420)

And because Python had an array,

Lex Fridman (10:06.140)

had two things that made me not filter it away.

Travis Oliphant (10:09.100)

Because I was filtering a bunch of stuff,

Lex Fridman (10:10.380)

as Yorick, I looked at Yorick,

Travis Oliphant (10:11.700)

I looked at a few other languages that are out there

Lex Fridman (10:14.420)

at the time in 1997, but it had arrays.

Travis Oliphant (10:17.700)

There's a library called Numeric

Lex Fridman (10:19.060)

that had just been written in 95,

Travis Oliphant (10:20.860)

like not very, not too much earlier.

Lex Fridman (10:23.740)

By an MIT alum, Jim Huganen.

Travis Oliphant (10:26.980)

You know, and I went back and read the mailing list

Lex Fridman (10:29.100)

to see the history of how it grew.

Lex Fridman (10:30.300)

And there was a very interesting,

Lex Fridman (10:31.220)

it's fascinating to do that actually,

Travis Oliphant (10:32.380)

to see how this emergent cooperation,

Lex Fridman (10:36.020)

unstructured cooperation happens in the open source world

Travis Oliphant (10:39.500)

that led to a lot of this collective programming,

Lex Fridman (10:43.300)

which is something maybe we might get into a little later,

Lex Fridman (10:45.140)

but what that looks like.

Lex Fridman (10:46.100)

What gap did Numeric fill?

Travis Oliphant (10:48.340)

Numeric filled the gap of having an array object.

Lex Fridman (10:50.260)

There was no array object.

Travis Oliphant (10:51.580)

There was no array.

Lex Fridman (10:52.420)

There was a one dimensional byte concept,

Lex Fridman (10:55.380)

but there was no n dimensional,

Lex Fridman (10:57.580)

two, three, four dimensional tensor they call it now.

Travis Oliphant (11:00.700)

I'm still in the category that a tensor is another thing

Lex Fridman (11:03.260)

and it's just an ndarray we should call it,

Lex Fridman (11:05.220)

but kind of lost that battle.

Lex Fridman (11:08.340)

There's many battles in this world,

Travis Oliphant (11:10.140)

some of which we win, some we lose.

Lex Fridman (11:12.060)

That's exactly right.

Travis Oliphant (11:13.620)

So, but it had no math to it.

Lex Fridman (11:17.180)

So Numeric had math and a basic way to think in arrays.

Lex Fridman (11:20.820)

So I was looking for that,

Lex Fridman (11:21.820)

and it had complex numbers,

Travis Oliphant (11:24.980)

a lot of programming languages.

Lex Fridman (11:26.380)

And you can see it because,

Travis Oliphant (11:28.100)

if you're just a computer scientist,

Lex Fridman (11:29.500)

you think, ah, complex numbers are just two floats.

Lex Fridman (11:32.060)

So you can, people can build that on.

Lex Fridman (11:34.980)

But in practice, a complex number

Travis Oliphant (11:36.740)

as one of the significant algebras

Lex Fridman (11:38.980)

that helps connect a lot of physical

Lex Fridman (11:40.740)

and mathematical ideas,

Lex Fridman (11:42.260)

particularly FFT for an electrical engineer.

Lex Fridman (11:45.100)

And it's a really important concept

Lex Fridman (11:48.160)

and not having it means you have to develop it

Travis Oliphant (11:50.820)

several times and those times may not share an approach.

Lex Fridman (11:54.300)

One of the common things in programming,

Travis Oliphant (11:55.700)

one of the things programming enables is abstractions.

Lex Fridman (11:59.060)

But when you have shared abstractions, it's even better.

Travis Oliphant (12:01.180)

It sort of gets to the level of language

Lex Fridman (12:02.980)

of actually we all think of this the same way,

Lex Fridman (12:05.540)

which is both powerful and dangerous, right?

Lex Fridman (12:07.940)

Because powerful in that we now can quickly

Travis Oliphant (12:11.180)

make bigger and higher level things

Lex Fridman (12:13.340)

on top of those abstractions dangerous

Travis Oliphant (12:14.800)

because it also limits us as to the things

Lex Fridman (12:17.100)

we maybe left behind in producing that abstraction,

Travis Oliphant (12:20.500)

which is at the heart of programming today

Lex Fridman (12:21.900)

and actually building around the programming world.

Travis Oliphant (12:24.420)

I think it's a fascinating philosophical topic.

Lex Fridman (12:26.580)

Yeah, they will continue for many years, I think.

Travis Oliphant (12:28.380)

They'll continue for many years.

Lex Fridman (12:29.220)

As we build more and more and more abstractions.

Travis Oliphant (12:31.260)

Yes, I often think about, you know,

Lex Fridman (12:32.340)

we have a world that's built on these abstractions

Lex Fridman (12:35.060)

that were they the only ones possible?

Lex Fridman (12:37.500)

Certainly not, but they led to,

Travis Oliphant (12:39.860)

you know, it's very hard to do it differently.

Lex Fridman (12:42.300)

Like there's an inertia that's very hard to,

Travis Oliphant (12:44.980)

you know, push out, push away from.

Lex Fridman (12:47.740)

That has implications for things like,

Travis Oliphant (12:49.640)

you know, the Julia language,

Lex Fridman (12:50.720)

which you have heard of, I'm sure.

Lex Fridman (12:52.680)

And I've met the creators and I liked Julia.

Lex Fridman (12:55.700)

It's a really cool language,

Lex Fridman (12:56.580)

but they struggled to kind of against the,

Lex Fridman (12:59.300)

just the tide of like this inertia of people using Python.

Travis Oliphant (13:03.420)

And, you know, there's strategies to approach that,

Lex Fridman (13:05.820)

but nonetheless, it's a phenomena.

Lex Fridman (13:07.580)

And sometimes, so I love complex numbers

Lex Fridman (13:09.580)

and I love to raise, so I looked at Python.

Lex Fridman (13:12.260)

And then I had the experience, I did some stuff in Python

Lex Fridman (13:15.260)

and I was just doing my PhD.

Lex Fridman (13:16.380)

So I was out, my focus was on,

Lex Fridman (13:19.700)

I was actually doing a combination of MRI and ultrasound

Lex Fridman (13:22.180)

and looking at a phenomenon called elastography,

Lex Fridman (13:24.740)

which is you push waves into the body

Lex Fridman (13:27.020)

and observe those waves, like you can actually measure them.

Lex Fridman (13:30.300)

And then you do mathematical inversion

Travis Oliphant (13:32.780)

to see what the elasticity is.

Lex Fridman (13:35.220)

And so that's the problem I was solving

Travis Oliphant (13:36.820)

is how to do that with both ultrasound and MRI.

Lex Fridman (13:39.780)

I needed some tool to do that with.

Lex Fridman (13:41.380)

So I was starting to use Python in 97.

Lex Fridman (13:44.260)

In 98, I went back, looked at what I'd written

Lex Fridman (13:47.340)

and realized I could still understand it,

Lex Fridman (13:49.560)

which is not the experience I'd had

Lex Fridman (13:50.900)

when doing Perl in 95, right?

Lex Fridman (13:53.660)

I'd done the same thing and then I looked back

Lex Fridman (13:55.620)

and I forgotten what I was even saying.

Lex Fridman (13:58.360)

Now, you know, I'm not saying, so that may,

Travis Oliphant (14:00.700)

hey, this may work, I like this.

Lex Fridman (14:02.400)

This is something I can retain

Travis Oliphant (14:04.980)

without becoming an expert per se.

Lex Fridman (14:07.620)

And so that led me to go, I'm gonna push more into this.

Lex Fridman (14:10.380)

And then that 98 was kind of when I started

Lex Fridman (14:14.820)

to fall in love with Python, I would say.

Travis Oliphant (14:18.300)

A few peculiar things about Python.

Lex Fridman (14:20.900)

So maybe compare it to Perl,

Travis Oliphant (14:22.940)

compare it to some of the other languages.

Lex Fridman (14:24.500)

So there's no braces.

Travis Oliphant (14:26.320)

Yeah.

Lex Fridman (14:27.160)

So space is used, indentation, I should say,

Travis Oliphant (14:31.960)

is used as part of the language.

Lex Fridman (14:33.980)

Yeah, right.

Lex Fridman (14:35.540)

So did you, I mean, that's quite a leap.

Lex Fridman (14:39.980)

Were you comfortable with that leap

Lex Fridman (14:41.180)

or were you just very open minded?

Lex Fridman (14:42.740)

It's a good question.

Travis Oliphant (14:43.580)

I was open minded, so I was cognizant of the concern.

Lex Fridman (14:48.040)

And it definitely has, it has specific challenges.

Travis Oliphant (14:52.060)

You know, cut and pasting.

Lex Fridman (14:53.520)

For example, when you're cut and pasting code,

Lex Fridman (14:55.460)

and if your editors aren't supportive of that,

Lex Fridman (14:57.220)

if you're putting it into a terminal,

Lex Fridman (14:58.980)

and particularly in the past when terminals

Lex Fridman (15:01.020)

didn't necessarily have the intelligence to manage it now.

Travis Oliphant (15:03.140)

Now, I, Python, and Jupyter Notebooks

Lex Fridman (15:05.100)

handle that just fine, so there's really no problem.

Lex Fridman (15:06.820)

But in the past, it created some challenges,

Lex Fridman (15:08.740)

formatting challenges, also mixed tabs and spaces.

Travis Oliphant (15:12.460)

If editors weren't, you weren't clear

Lex Fridman (15:14.740)

on what was happening, you would have these issues.

Lex Fridman (15:16.860)

So there were really concrete reasons about it

Lex Fridman (15:19.180)

that I heard and understood.

Travis Oliphant (15:20.400)

I never really encountered a problem with it personally.

Lex Fridman (15:23.960)

Like, it was occasional annoyances,

Lex Fridman (15:26.480)

but I really liked the fact

Lex Fridman (15:28.420)

that it didn't have all this extra characters, right?

Travis Oliphant (15:31.060)

That these extra characters didn't show up

Lex Fridman (15:33.100)

in my visual field when I was just trying

Travis Oliphant (15:35.420)

to process understanding a snippet of code.

Lex Fridman (15:38.000)

Yeah, there's a cleanness to it.

Travis Oliphant (15:39.260)

But, I mean, the idea is supposed to be

Lex Fridman (15:41.140)

that Perl also has a cleanness to it

Travis Oliphant (15:43.300)

because of the minimalism of how many characters

Lex Fridman (15:46.500)

it takes to express a certain thing.

Lex Fridman (15:48.420)

So it's very compact.

Lex Fridman (15:49.820)

But what you realize with that compactness comes,

Travis Oliphant (15:53.560)

there's a culture that prizes compactness,

Lex Fridman (15:57.100)

and so the code gets more and more compact

Lex Fridman (15:58.900)

and less and less readable to a point where it's like,

Lex Fridman (16:03.600)

like, to be a good programmer in Perl,

Travis Oliphant (16:05.420)

you write code that's basically unreadable.

Lex Fridman (16:07.820)

There's a culture, like.

Travis Oliphant (16:09.100)

Correct, and you're proud of it.

Lex Fridman (16:10.860)

Yeah, you're proud of it.

Travis Oliphant (16:12.460)

Right, exactly, and it's like, feels good.

Lex Fridman (16:14.140)

And it's really selective.

Travis Oliphant (16:16.660)

It means you have to be an expert in Perl to understand it.

Lex Fridman (16:20.380)

Whereas Python allowed you not to have to be an expert.

Travis Oliphant (16:22.980)

You didn't have to take all this brain energy.

Lex Fridman (16:24.740)

You could leverage, what I say,

Travis Oliphant (16:25.660)

you could leverage your English language center,

Lex Fridman (16:28.180)

which you're using all the time.

Travis Oliphant (16:29.980)

I've wondered about other languages,

Lex Fridman (16:31.180)

particularly non Latin based languages.

Travis Oliphant (16:34.680)

Latin based languages with the characters are at least similar.

Lex Fridman (16:37.220)

I think people have an easier time,

Lex Fridman (16:38.620)

but I don't know what it's like to be a Japanese

Lex Fridman (16:41.300)

or a Chinese person trying to learn different syntax.

Lex Fridman (16:46.900)

Like, what would computer programming look like in that?

Lex Fridman (16:49.740)

I haven't looked at that at all,

Lex Fridman (16:50.780)

but it certainly doesn't,

Lex Fridman (16:52.020)

you know, leveraging your Chinese language center,

Travis Oliphant (16:54.300)

I'm not sure Python or any programming does that.

Lex Fridman (16:57.060)

But that was a big deal.

Travis Oliphant (16:58.140)

The fact that it was accessible, I could be a scientist.

Lex Fridman (17:00.340)

What I really liked is many programming languages

Travis Oliphant (17:02.900)

really demand a lot of you, and you can get a lot,

Lex Fridman (17:04.740)

you know, you do a lot if you learn it.

Lex Fridman (17:07.200)

But Python enables you to do a lot

Lex Fridman (17:08.900)

without demanding a lot of you.

Travis Oliphant (17:11.180)

There's nuance to that statement,

Lex Fridman (17:13.100)

but it certainly was, it's more accessible.

Lex Fridman (17:15.340)

So more people could actually, as a scientist,

Lex Fridman (17:18.040)

as somebody who, or an engineer,

Travis Oliphant (17:19.860)

who was trying to solve another problem

Lex Fridman (17:21.460)

besides point programming,

Travis Oliphant (17:23.300)

I could still use this language and get things done

Lex Fridman (17:26.000)

and be happy about it.

Lex Fridman (17:27.340)

And I was also comfortable in C at that time.

Lex Fridman (17:30.100)

And MATLAB, you did a little bit of that.

Lex Fridman (17:31.340)

And MATLAB, I did a lot before that, exactly.

Lex Fridman (17:33.180)

So I was comfortable in,

Travis Oliphant (17:34.900)

those three languages were really the tools I used

Lex Fridman (17:37.580)

during my studies and schooling.

Lex Fridman (17:40.540)

But to your point about language helping you think,

Lex Fridman (17:42.620)

one of the big things about MATLAB was it was,

Lex Fridman (17:44.580)

and APL before it, I don't know if you remember APL.

Lex Fridman (17:48.300)

APL is actually the predecessor of array based programming,

Travis Oliphant (17:51.660)

which I think is really an underappreciated,

Lex Fridman (17:54.160)

if I talk to people who are just steeped

Travis Oliphant (17:55.340)

in computer programming, computer science,

Lex Fridman (17:57.640)

like most of the people that Microsoft has hired

Travis Oliphant (17:59.460)

in the past, for example,

Lex Fridman (18:01.140)

Microsoft as a company generally did not understand

Travis Oliphant (18:03.900)

array based programming.

Lex Fridman (18:05.220)

Like culturally, they didn't understand it.

Lex Fridman (18:06.620)

So they kept missing the boat,

Lex Fridman (18:08.560)

kept missing the understanding of what this was.

Travis Oliphant (18:11.580)

They've gotten better,

Lex Fridman (18:12.740)

but there's still a whole culture of folks

Travis Oliphant (18:14.420)

that doesn't, programming, that's systems programming

Lex Fridman (18:17.980)

or web programming or lists and maps.

Lex Fridman (18:20.380)

And what about an n dimensional array?

Lex Fridman (18:22.520)

Oh yeah, that's just an implementation detail.

Travis Oliphant (18:24.700)

Well, you can think that,

Lex Fridman (18:26.700)

but then actually if you have that as a construct,

Travis Oliphant (18:28.800)

you actually think differently.

Lex Fridman (18:29.860)

APL was the first language to understand that.

Lex Fridman (18:31.660)

And it was in the sixties, right?

Lex Fridman (18:33.460)

The challenge of APL is APL had very dense,

Travis Oliphant (18:36.780)

not only glyphs, like new characters, new glyphs,

Lex Fridman (18:39.340)

but they even had a new keyboard

Travis Oliphant (18:40.480)

because to produce those glyphs,

Lex Fridman (18:42.340)

this is back in the early days in computing

Travis Oliphant (18:43.980)

when the QWERTY keyboard maybe wasn't as established,

Lex Fridman (18:47.980)

like, well, we can have a new keyboard, no big deal.

Lex Fridman (18:50.780)

But it was a big deal and it didn't catch on.

Lex Fridman (18:52.860)

And the language APL, very much like Perl,

Travis Oliphant (18:56.500)

as people would pride themselves on how much,

Lex Fridman (18:58.620)

could they write the game of life

Travis Oliphant (18:59.740)

in 30 characters of APL.

Lex Fridman (19:03.100)

APL has characters that mean summation

Lex Fridman (19:06.060)

and they have adverbs,

Lex Fridman (19:08.180)

they would have adjectives and these things called adverbs,

Travis Oliphant (19:10.060)

which are like methods, like reduction,

Lex Fridman (19:12.220)

reduction would be an adverb on an ad operator, right?

Travis Oliphant (19:15.320)

So, but doing, using these tools you could construct

Lex Fridman (19:18.660)

and then you start to think at that level,

Travis Oliphant (19:20.880)

you think in n dimensions is something I like to say,

Lex Fridman (19:22.900)

and you start to think differently about data at that point.

Travis Oliphant (19:25.500)

Now you're, it really helps.

Lex Fridman (19:27.500)

Yeah, I mean, outside of programming,

Travis Oliphant (19:30.100)

if you really internalize linear algebra as a course,

Lex Fridman (19:33.700)

I mean, it's philosophically allows you

Travis Oliphant (19:35.580)

to think of the world differently.

Lex Fridman (19:37.220)

It's almost like liberating, you don't have to,

Travis Oliphant (19:39.700)

you don't have to think about the individual numbers

Lex Fridman (19:42.100)

in the n dimensional array.

Travis Oliphant (19:44.240)

You could think of it as an object in itself

Lex Fridman (19:46.140)

and all of a sudden this world can open up.

Travis Oliphant (19:48.500)

You're saying MATLAB and APL were like the early C,

Lex Fridman (19:52.660)

I don't know if many languages got that right ever.

Travis Oliphant (19:54.980)

No, no, no they didn't.

Lex Fridman (19:56.860)

Even still.

Travis Oliphant (19:57.700)

Even still, I would say.

Lex Fridman (19:58.820)

I mean, NumPy is an inheritor of the traditions

Lex Fridman (1:00:01.900)

but advanced indexing so that you could do masked indexing

Lex Fridman (1:00:06.500)

and indirect indexing instead of just slicing.

Lex Fridman (1:00:09.940)

So for people who don't know, and maybe you can elaborate,

Lex Fridman (1:00:13.020)

NumPy, I guess the vision in the narrowest sense

Travis Oliphant (1:00:17.660)

is to have this object that represents

Lex Fridman (1:00:21.460)

n dimensional arrays.

Lex Fridman (1:00:23.180)

And like at any level of abstraction you want,

Lex Fridman (1:00:26.300)

but basically it could be a black box

Travis Oliphant (1:00:28.220)

that you can investigate in ways that you would naturally

Lex Fridman (1:00:30.940)

want to investigate such objects.

Travis Oliphant (1:00:33.340)

Yes, exactly.

Lex Fridman (1:00:34.180)

So you could do math on it easily.

Travis Oliphant (1:00:35.740)

Math on it easily, yeah.

Lex Fridman (1:00:37.180)

So it had an associated library of math operations

Lex Fridman (1:00:39.860)

and effectively SciPy became an even larger operate set

Lex Fridman (1:00:43.220)

of math operations.

Lex Fridman (1:00:44.940)

So the key for me was I was going to write NumPy

Lex Fridman (1:00:48.020)

and then move SciPy to depend on NumPy.

Travis Oliphant (1:00:50.340)

In fact, early on, one of the initial proposals

Lex Fridman (1:00:52.980)

was that we would just write SciPy

Lex Fridman (1:00:54.540)

and it would have the numeric object inside of it.

Lex Fridman (1:00:56.660)

And it'd be SciPy.array or something.

Travis Oliphant (1:00:59.780)

That turned out to be problematic because numeric

Lex Fridman (1:01:02.180)

already had a little mini library of linear algebra

Lex Fridman (1:01:04.820)

and some functions, and it had enough momentum,

Lex Fridman (1:01:08.020)

enough users that nobody wanted to,

Travis Oliphant (1:01:10.340)

they wanted backward compatibility.

Lex Fridman (1:01:12.060)

One of the big challenges of NumPy

Travis Oliphant (1:01:13.740)

was I had to be backward compatible

Lex Fridman (1:01:14.980)

with both numeric and NumArray

Travis Oliphant (1:01:16.980)

in order to allow both of those communities to come together.

Lex Fridman (1:01:19.300)

There was a ton of work in creating

Travis Oliphant (1:01:21.140)

that backward compatibility

Lex Fridman (1:01:22.580)

that also created echoes in today's object.

Travis Oliphant (1:01:25.420)

Like some of the complexity in today's object

Lex Fridman (1:01:27.180)

is actually from that goal of backward compatibility

Travis Oliphant (1:01:30.060)

to these other communities,

Lex Fridman (1:01:31.380)

which if you didn't have that, you'd do something different,

Travis Oliphant (1:01:34.620)

which is instructive because a lot of things are there.

Lex Fridman (1:01:37.740)

You think, what is that there for?

Travis Oliphant (1:01:38.940)

It's like, well, it's a remnant.

Lex Fridman (1:01:41.380)

It's an artifact of its historical existence.

Travis Oliphant (1:01:45.220)

By the way, I love the empathy

Lex Fridman (1:01:46.780)

and the lack of ego behind that

Travis Oliphant (1:01:48.460)

because I feel, you see that in the split

Lex Fridman (1:01:51.420)

in the JavaScript framework, for example,

Travis Oliphant (1:01:53.340)

the arbitrary branching.

Lex Fridman (1:01:54.860)

Right.

Travis Oliphant (1:01:56.980)

I think in order to unite people,

Lex Fridman (1:01:59.020)

you have to kind of put your ego aside

Lex Fridman (1:02:00.620)

and truly listen to others.

Lex Fridman (1:02:02.260)

You do.

Lex Fridman (1:02:03.100)

What do you love about NumArray?

Lex Fridman (1:02:04.820)

What do you love about Numeric?

Travis Oliphant (1:02:06.020)

Like actually get a sense,

Lex Fridman (1:02:07.460)

we were talking about languages earlier,

Travis Oliphant (1:02:08.860)

sort of empathize to the culture,

Lex Fridman (1:02:11.100)

the people that love something about this particular API,

Travis Oliphant (1:02:14.660)

some of the naming style

Lex Fridman (1:02:18.100)

or the actual usage patterns

Lex Fridman (1:02:21.220)

and truly understand them

Lex Fridman (1:02:22.820)

and so that you can create that same draw

Travis Oliphant (1:02:26.780)

in the united thing. I completely agree.

Lex Fridman (1:02:28.620)

I completely agree.

Lex Fridman (1:02:29.460)

And you have to also have enough passion

Lex Fridman (1:02:31.780)

that you'll do it.

Travis Oliphant (1:02:32.620)

It can't be just like a perfunctory,

Lex Fridman (1:02:34.660)

oh yes, I'll listen to you

Lex Fridman (1:02:36.500)

and then I'm not really that excited about it.

Lex Fridman (1:02:38.380)

So it really is an aspect,

Travis Oliphant (1:02:39.620)

it's a philosophical, like there's a philia,

Lex Fridman (1:02:42.260)

there's a love of esteeming of others.

Travis Oliphant (1:02:44.260)

It's actually at the heart of what,

Lex Fridman (1:02:47.060)

it's sort of a life philosophy for me, right?

Travis Oliphant (1:02:49.220)

That I'm constantly pursuing and that helped,

Lex Fridman (1:02:51.540)

absolutely helped.

Travis Oliphant (1:02:52.660)

Makes me wonder in a philosophical,

Lex Fridman (1:02:54.260)

like looking at human civilization as one object,

Travis Oliphant (1:02:57.460)

it makes me wonder how we can copy and paste Travis's

Lex Fridman (1:02:59.980)

in this book.

Travis Oliphant (1:03:00.820)

Well, some aspects, maybe.

Lex Fridman (1:03:03.300)

Some aspects, right, right, exactly.

Travis Oliphant (1:03:05.220)

Well, it's a good question.

Lex Fridman (1:03:07.300)

How do we teach this?

Lex Fridman (1:03:08.140)

How do we encourage it?

Lex Fridman (1:03:09.300)

How do we lift it?

Travis Oliphant (1:03:10.140)

Because so much of the software world,

Lex Fridman (1:03:12.700)

it's giant communities, right?

Lex Fridman (1:03:15.140)

But it seems like so much is moved by,

Lex Fridman (1:03:16.820)

like little individuals.

Travis Oliphant (1:03:18.180)

You talk about like Linus Torvalds.

Lex Fridman (1:03:21.020)

It's like, could you have not,

Lex Fridman (1:03:23.380)

could you have had Linux without him?

Lex Fridman (1:03:25.980)

Could you?

Travis Oliphant (1:03:26.820)

Yeah, Guido and Python.

Lex Fridman (1:03:28.140)

Guido and Python.

Travis Oliphant (1:03:28.980)

Guido and Python.

Lex Fridman (1:03:29.820)

Well, the iPy community particularly,

Travis Oliphant (1:03:30.980)

it's like I said, we wanted to build this big thing,

Lex Fridman (1:03:32.820)

but ultimately we didn't.

Lex Fridman (1:03:33.780)

What happened is we had Mavericks and champions

Lex Fridman (1:03:36.060)

like John Hunter who created Matplotlib.

Travis Oliphant (1:03:37.780)

We had Fernando Perez who created iPython.

Lex Fridman (1:03:39.940)

And so we sort of inspired each other,

Lex Fridman (1:03:42.260)

but then it kind of, there's sort of a culture

Lex Fridman (1:03:43.980)

of this selfless giving, the stewardship mentality,

Travis Oliphant (1:03:47.820)

as opposed to ownership mentality,

Lex Fridman (1:03:49.140)

but stewardship and community focused,

Travis Oliphant (1:03:54.040)

community focused, but intentional work.

Lex Fridman (1:03:56.620)

Like not waiting for everybody else to do the work,

Lex Fridman (1:03:58.900)

but you're doing it for the benefit of others

Lex Fridman (1:04:00.700)

and not worried about what you're gonna get.

Travis Oliphant (1:04:04.020)

You're not worried about the credit.

Lex Fridman (1:04:04.860)

You're not worried about what you're gonna get.

Travis Oliphant (1:04:05.860)

You're worried about, I later realized

Lex Fridman (1:04:07.580)

that I have to worry a little about credit,

Travis Oliphant (1:04:09.000)

not because I want the credit,

Lex Fridman (1:04:10.300)

because I want people to understand

Lex Fridman (1:04:11.380)

what led to the results.

Lex Fridman (1:04:13.020)

Like, I don't, it's not about me.

Travis Oliphant (1:04:15.060)

It's I want to understand this is what led to the result.

Lex Fridman (1:04:17.540)

So let's like, I think doing,

Lex Fridman (1:04:18.980)

and this is what had no impact on the result.

Lex Fridman (1:04:21.100)

Like let's promote, just like you said,

Travis Oliphant (1:04:23.420)

I want to promote the attributes

Lex Fridman (1:04:25.100)

that help make us better off.

Lex Fridman (1:04:26.520)

How do we make more of West McKinney?

Lex Fridman (1:04:28.820)

Like West McKinney was critical to the success of Python

Travis Oliphant (1:04:31.620)

because of his creation of pandas,

Lex Fridman (1:04:33.420)

which is the roots of that were all the way back

Travis Oliphant (1:04:36.420)

in numeric and num array and numpy,

Lex Fridman (1:04:40.260)

where numpy created an array of records.

Travis Oliphant (1:04:43.180)

West started to use that almost like a data frame,

Lex Fridman (1:04:45.980)

except it's an array of records.

Lex Fridman (1:04:47.840)

And data frame, the challenge is,

Lex Fridman (1:04:49.780)

okay, if you want to augment it at another column,

Travis Oliphant (1:04:52.240)

you have to insert, you have to do all this memory movement

Lex Fridman (1:04:54.700)

to insert a column.

Travis Oliphant (1:04:55.660)

Whereas data frames became,

Lex Fridman (1:04:57.180)

oh, I'm going to have a loose collection of arrays.

Lex Fridman (1:05:00.460)

So it's a record of arrays that is a part of a data frame.

Lex Fridman (1:05:03.980)

And we thought about that back in the memory days,

Lex Fridman (1:05:05.560)

but West ended up doing the work to build it.

Lex Fridman (1:05:08.940)

And then also the operations that were relevant

Travis Oliphant (1:05:11.300)

for data processing.

Lex Fridman (1:05:12.620)

What I noticed is just that each of these little things

Travis Oliphant (1:05:15.220)

creates just another tick, another up.

Lex Fridman (1:05:17.380)

So numpy ultimately took a little while,

Travis Oliphant (1:05:19.940)

about six months in, people started to join me,

Lex Fridman (1:05:22.700)

Francesc Altad, Robert Kern, Charles Harris.

Lex Fridman (1:05:27.300)

And these people are many of the unsung heroes, I would say.

Lex Fridman (1:05:30.300)

People who are, you know,

Travis Oliphant (1:05:31.980)

they sometimes don't get the credit they deserve

Lex Fridman (1:05:34.100)

because they were critical both to support,

Travis Oliphant (1:05:36.540)

like, you know, it's hard and you want,

Lex Fridman (1:05:38.260)

you need some support, people need support.

Lex Fridman (1:05:40.340)

And I needed just encouragement.

Lex Fridman (1:05:41.580)

And they were helping and encouraged by contributing.

Lex Fridman (1:05:43.860)

And once, the big thing for me was when John Hunter,

Lex Fridman (1:05:48.240)

he had previously done kind of a simple thing

Travis Oliphant (1:05:50.180)

called numerics to kind of, you know, between numeric

Lex Fridman (1:05:52.820)

and numerae, he had a little high level tool

Travis Oliphant (1:05:55.100)

that would just select each one for matplotlib.

Lex Fridman (1:05:57.900)

In 2006, he finally said,

Travis Oliphant (1:06:00.420)

we're gonna just make numpy the dependency of matplotlib.

Lex Fridman (1:06:03.220)

As soon as he did that,

Lex Fridman (1:06:04.420)

and I remember specifically when he did that,

Lex Fridman (1:06:06.100)

I said, okay, we've done it.

Travis Oliphant (1:06:07.900)

Like, that was when I knew we had to see success.

Lex Fridman (1:06:11.260)

Before then it was still unsure,

Lex Fridman (1:06:13.620)

but that kind of started a roller coaster.

Lex Fridman (1:06:15.060)

And then 2006 to 2009.

Lex Fridman (1:06:17.900)

And then I've been floored by what it's done.

Lex Fridman (1:06:20.940)

Like, I knew it would help.

Travis Oliphant (1:06:22.900)

I had no idea how much it would help.

Lex Fridman (1:06:25.380)

Right, so.

Lex Fridman (1:06:26.300)

And it has to do with, again, the language thing.

Lex Fridman (1:06:28.660)

It just, people started to think in terms of numpy.

Travis Oliphant (1:06:31.940)

Yes.

Lex Fridman (1:06:32.820)

And that opened up a whole new way of thinking.

Lex Fridman (1:06:36.460)

And part of the story that you kind of mentioned,

Lex Fridman (1:06:39.220)

but maybe you can elaborate,

Travis Oliphant (1:06:42.980)

is it seems like at some point in the story,

Lex Fridman (1:06:46.320)

Python took over science and data science.

Travis Oliphant (1:06:50.800)

Yes.

Lex Fridman (1:06:51.640)

And bigger than that,

Travis Oliphant (1:06:54.800)

the scientific community started to think like programmers

Lex Fridman (1:07:00.160)

or started to utilize the tools of computers to do,

Travis Oliphant (1:07:04.280)

like at a scale that wasn't done with Fortran.

Lex Fridman (1:07:06.640)

Like at this gigantic scale,

Travis Oliphant (1:07:09.320)

they started to open in their heart.

Lex Fridman (1:07:10.760)

And then Python was the thing.

Travis Oliphant (1:07:12.040)

I mean, there's a few other competitors, I guess,

Lex Fridman (1:07:14.280)

but Python, I think, really, really took over.

Travis Oliphant (1:07:16.960)

I agree.

Lex Fridman (1:07:17.800)

There's a lot of stories here

Travis Oliphant (1:07:18.620)

that are kind of during this journey,

Lex Fridman (1:07:19.720)

because this is sort of the start of this journey in 2005, 2006.

Lex Fridman (1:07:23.240)

So my tenure committee, I applied for tenure in 2006, 2007.

Lex Fridman (1:07:28.180)

It came back, I split the department.

Travis Oliphant (1:07:29.780)

I was very polarizing.

Lex Fridman (1:07:31.300)

I had some huge fans

Lex Fridman (1:07:32.560)

and then some people that said no way, right?

Lex Fridman (1:07:34.380)

So it was very, I was a polarizing figure in the department.

Travis Oliphant (1:07:36.840)

It went all the way up to the university president.

Lex Fridman (1:07:39.800)

Ultimately, my department chair had the sway

Lex Fridman (1:07:42.760)

and they didn't say no.

Lex Fridman (1:07:43.760)

They said, come back in two years and do it again.

Lex Fridman (1:07:46.360)

And I went, eh, at that point, I was like,

Lex Fridman (1:07:49.680)

I mean, I had this interest in entrepreneurship,

Travis Oliphant (1:07:52.840)

this interest in not the academic circles,

Lex Fridman (1:07:56.400)

not the, like, how do we make industry work?

Lex Fridman (1:07:59.680)

So I do have to give credit to that exploration of economics

Lex Fridman (1:08:03.060)

because that led me, oh, I had a lot of opinions.

Travis Oliphant (1:08:06.540)

I was actually very libertarian at the time.

Lex Fridman (1:08:09.520)

And I still have some libertarian trends,

Lex Fridman (1:08:11.840)

but I'm more of a, I'm more of a collectivist libertarian.

Lex Fridman (1:08:15.880)

So you value broadly, philosophically freedom.

Travis Oliphant (1:08:18.720)

I value broadly, philosophically freedom,

Lex Fridman (1:08:20.360)

but I also understand the power of communities,

Travis Oliphant (1:08:23.440)

like the power of collective behavior.

Lex Fridman (1:08:26.200)

And so what's that balance, right?

Travis Oliphant (1:08:27.840)

That makes sense.

Lex Fridman (1:08:29.800)

So by the time I was just,

Travis Oliphant (1:08:31.520)

I gotta go out and explore this entrepreneur world.

Lex Fridman (1:08:33.380)

So I left academia.

Travis Oliphant (1:08:34.220)

I said, no thanks, called my friend, Eric, here,

Lex Fridman (1:08:37.820)

who had, his company was going.

Lex Fridman (1:08:39.560)

I said, hey, could I join you and start this trend?

Lex Fridman (1:08:43.120)

And he, at that time they were using SciFi a lot.

Travis Oliphant (1:08:45.920)

They were trying to get clients.

Lex Fridman (1:08:47.120)

And so I came down to Texas.

Lex Fridman (1:08:48.760)

And in Texas is where I sort of,

Lex Fridman (1:08:51.160)

it's my entrepreneur world, right?

Travis Oliphant (1:08:53.440)

I left academia and went to entrepreneur world in 2007.

Lex Fridman (1:08:57.360)

So I moved here in 2007, kind of took a leap,

Travis Oliphant (1:08:59.920)

knew nothing really about business,

Lex Fridman (1:09:01.600)

knew nothing about a lot of stuff there.

Travis Oliphant (1:09:05.100)

There's, you know, for a long time,

Lex Fridman (1:09:06.980)

I've kept some connections to a lot of academics

Travis Oliphant (1:09:08.980)

because I still value it.

Lex Fridman (1:09:10.080)

I still love the scientific tradition.

Travis Oliphant (1:09:12.520)

I still value the essence and the soul and the heart

Lex Fridman (1:09:15.240)

of what is possible.

Travis Oliphant (1:09:17.320)

Don't like a lot of the administration

Lex Fridman (1:09:21.380)

and the kind of, we can go into detail about why

Lex Fridman (1:09:24.160)

and where and how this happens,

Lex Fridman (1:09:25.320)

what are some of the challenges.

Travis Oliphant (1:09:26.520)

I don't know, but I'm with you.

Lex Fridman (1:09:28.480)

So I'm still affiliated with MIT.

Travis Oliphant (1:09:31.840)

I still love MIT because there's magic there.

Lex Fridman (1:09:35.600)

There's people I talk to, like researchers, faculty,

Travis Oliphant (1:09:40.320)

in those conversations and the whiteboard

Lex Fridman (1:09:43.120)

and just the conversation, that's magic there.

Travis Oliphant (1:09:46.220)

All the other stuff, the administration,

Lex Fridman (1:09:48.120)

all that kind of stuff seems to,

Travis Oliphant (1:09:52.020)

you don't wanna say too harshly criticize

Lex Fridman (1:09:54.920)

sort of bureaucracies, but there's a lag

Travis Oliphant (1:09:57.680)

that seems to get in the way of the magic.

Lex Fridman (1:10:00.800)

And I'm still have a lot of hope

Travis Oliphant (1:10:03.800)

that that can change because I don't often see

Lex Fridman (1:10:08.320)

that particular type of magic elsewhere in the industry.

Lex Fridman (1:10:12.840)

So like we need that and we need that flame going.

Lex Fridman (1:10:15.800)

And it's the same thing as exactly as you said,

Travis Oliphant (1:10:19.120)

it has the same kind of elements

Lex Fridman (1:10:20.560)

like the open source community does.

Travis Oliphant (1:10:23.240)

And, but then if you, like the reason I stepped away,

Lex Fridman (1:10:27.160)

the reason I'm here, just like you did in Austin is like,

Travis Oliphant (1:10:30.260)

if I wanna build one robot, I'll stay at MIT.

Lex Fridman (1:10:33.240)

But if I wanna build millions and make money enough

Travis Oliphant (1:10:37.460)

to where I can explore the magic of that, then you can't.

Lex Fridman (1:10:41.000)

And I think that dance is...

Lex Fridman (1:10:44.160)

That translational dance has been lost a bit, right?

Lex Fridman (1:10:47.480)

And there's a lot of reasons for that.

Travis Oliphant (1:10:48.640)

I'm not, I'm certainly not an expert on this stuff.

Lex Fridman (1:10:50.160)

I can opine like anybody else,

Lex Fridman (1:10:51.660)

but I realized that I wanted to explore entrepreneurship,

Lex Fridman (1:10:55.820)

which I, and really figure out,

Lex Fridman (1:10:57.720)

and it's been a driving passion for 20 years, 25 years.

Lex Fridman (1:11:01.560)

How do we connect capital markets and company?

Travis Oliphant (1:11:06.480)

Cause again, I fell in love with the notion of,

Lex Fridman (1:11:07.880)

oh, profit seeking on its own is not a bad thing.

Travis Oliphant (1:11:11.160)

It's actually a coordination mechanism

Lex Fridman (1:11:13.520)

for allocating resources that, you know,

Lex Fridman (1:11:16.480)

in an emergent way, right?

Lex Fridman (1:11:18.000)

That respects everybody's opinions, right?

Lex Fridman (1:11:20.720)

So this is actually powerful.

Lex Fridman (1:11:21.880)

So I say all the time, when I make a company

Lex Fridman (1:11:25.320)

and we do something that makes profit,

Lex Fridman (1:11:27.260)

what we're saying is, hey,

Travis Oliphant (1:11:28.100)

we're collecting of the world's resources

Lex Fridman (1:11:29.800)

and voluntarily people are asking us

Travis Oliphant (1:11:31.480)

to do something that they like.

Lex Fridman (1:11:33.000)

And that's a huge deal.

Lex Fridman (1:11:34.000)

And so I really liked that energy.

Lex Fridman (1:11:36.120)

So that's what I came to do and to learn

Lex Fridman (1:11:37.560)

and to try to figure out.

Lex Fridman (1:11:38.480)

And that's what I've been kind of stumbling through

Travis Oliphant (1:11:40.120)

since for the past 14 years.

Lex Fridman (1:11:40.960)

And that's 2007.

Travis Oliphant (1:11:42.580)

2007, yeah.

Lex Fridman (1:11:43.420)

And so you were still working at NoPi.

Lex Fridman (1:11:44.960)

So NoPi was just emerging.

Lex Fridman (1:11:46.560)

Just emerging.

Travis Oliphant (1:11:47.400)

One of the things I've done,

Lex Fridman (1:11:49.160)

it's worth mentioning because it emphasizes

Travis Oliphant (1:11:51.480)

the exploratory nature of my thinking at the time.

Lex Fridman (1:11:53.840)

I said, well, I don't know how to fund this thing.

Travis Oliphant (1:11:55.240)

I've got a graduate student I'm paying for

Lex Fridman (1:11:56.720)

and I've got no funding for him.

Lex Fridman (1:11:57.880)

And I had done some fundraising from the public

Lex Fridman (1:12:00.520)

to try to get public fundraisers in my lab.

Travis Oliphant (1:12:02.800)

I didn't really wanna go out

Lex Fridman (1:12:03.880)

and just do the fundraising circuit

Travis Oliphant (1:12:05.360)

the way it's traditionally done.

Lex Fridman (1:12:06.920)

So I wrote a book and I said, I'm gonna write a book

Lex Fridman (1:12:09.960)

and I'm gonna charge for it.

Lex Fridman (1:12:11.440)

It was called Guide to NoPi.

Lex Fridman (1:12:12.720)

And so ultimately NoPi became

Lex Fridman (1:12:14.040)

documentation driven development

Travis Oliphant (1:12:15.960)

because I basically wrote the book

Lex Fridman (1:12:17.280)

and made sure the stuff worked or the book would work.

Lex Fridman (1:12:19.760)

So it really helped actually make NoPi become a thing.

Lex Fridman (1:12:23.040)

So writing that book,

Lex Fridman (1:12:25.800)

and it's not a page turner.

Lex Fridman (1:12:28.200)

Guide to NoPi is not a book you pick up

Lex Fridman (1:12:29.680)

and go, oh, this is great, over the fire.

Lex Fridman (1:12:31.520)

But it's where you could find the details,

Travis Oliphant (1:12:33.640)

like how'd all this work.

Lex Fridman (1:12:34.720)

And a lot of people love that book.

Lex Fridman (1:12:36.520)

And so a lot of people ended up,

Lex Fridman (1:12:38.040)

so I said, look, I need to, so I'm gonna charge for it.

Lex Fridman (1:12:41.600)

And I got some flack for that.

Lex Fridman (1:12:42.760)

Not that much, just probably five angry messages,

Travis Oliphant (1:12:45.920)

people yelling at me saying I was a bad guy

Lex Fridman (1:12:49.960)

for charging for this book.

Lex Fridman (1:12:51.360)

Was one of them Richard Stallman?

Lex Fridman (1:12:53.280)

No. Just kidding.

Travis Oliphant (1:12:54.120)

No, I haven't really had any interaction with him personally,

Lex Fridman (1:12:56.920)

like I said, but there were a few,

Lex Fridman (1:12:59.840)

but actually surprisingly not.

Lex Fridman (1:13:01.280)

There was actually a lot of people like,

Travis Oliphant (1:13:02.760)

no, it's fine, you can charge for a book.

Lex Fridman (1:13:04.240)

That's no big deal.

Travis Oliphant (1:13:05.080)

We know that's a way you can try to make money

Lex Fridman (1:13:07.080)

around open source.

Lex Fridman (1:13:07.920)

So what I did, I did it in an interesting way.

Lex Fridman (1:13:10.160)

I said, well, kind of my ideas around IP law and stuff.

Travis Oliphant (1:13:14.280)

I love the idea you can share something, you can spread it.

Lex Fridman (1:13:16.120)

Like once it's, the fact that you have a thing

Lex Fridman (1:13:18.280)

and copying is free, but the creation is not free.

Lex Fridman (1:13:21.640)

So how do you fund the creation and allow the copying?

Lex Fridman (1:13:25.600)

And in software, it's a little more complicated than that

Lex Fridman (1:13:27.040)

because creation is actually a continuous thing.

Travis Oliphant (1:13:29.360)

It's not like you build a widget and it's done.

Lex Fridman (1:13:31.160)

It's sort of a process of emerging

Lex Fridman (1:13:32.640)

and continuing to create.

Lex Fridman (1:13:34.560)

But I wrote the book

Lex Fridman (1:13:35.520)

and had this market determined price thing.

Lex Fridman (1:13:37.520)

I said, look, I need, I think I said 250,000.

Travis Oliphant (1:13:40.720)

If I make 250,000 from this book, I'll make it free.

Lex Fridman (1:13:44.280)

So as soon as I get that much money,

Travis Oliphant (1:13:45.760)

or I said five years, so there's a time limit.

Lex Fridman (1:13:48.960)

Like it's not forever.

Travis Oliphant (1:13:49.800)

That's really cool.

Lex Fridman (1:13:50.640)

It's amazing.

Travis Oliphant (1:13:51.680)

I released it on this.

Lex Fridman (1:13:53.080)

And it's actually interesting

Travis Oliphant (1:13:54.240)

because one of the people

Lex Fridman (1:13:55.800)

who also thought that was interesting

Travis Oliphant (1:13:57.040)

ended up being Chris White,

Lex Fridman (1:13:58.600)

who was the director of DARPA project

Travis Oliphant (1:14:01.360)

that we got funding through at Anaconda.

Lex Fridman (1:14:02.920)

And the reason he even called us back

Travis Oliphant (1:14:04.640)

is because he remembered my name from this book

Lex Fridman (1:14:06.720)

and he thought that was interesting.

Lex Fridman (1:14:08.080)

And so even though we hadn't gone to the demo days,

Lex Fridman (1:14:10.880)

we applied and the people said, yeah,

Travis Oliphant (1:14:12.680)

nobody ever gets this without coming to the demo day first.

Lex Fridman (1:14:15.360)

This is the first time I've seen it.

Lex Fridman (1:14:16.320)

But it's because I knew, you know,

Lex Fridman (1:14:18.200)

Chris had done this and had this interaction.

Lex Fridman (1:14:19.640)

So it did have impact.

Lex Fridman (1:14:21.680)

I was actually really, really pleased by the result.

Travis Oliphant (1:14:23.880)

I mean, I ended up in three years, I made 90,000.

Lex Fridman (1:14:27.360)

So sold 30,000 copies by myself.

Travis Oliphant (1:14:29.480)

I just put it up on, you know, use PayPal and sold it.

Lex Fridman (1:14:33.000)

And that was my first taste of kind of, okay,

Travis Oliphant (1:14:36.040)

this can work to some degree.

Lex Fridman (1:14:37.600)

And I, you know, all over the world, right?

Travis Oliphant (1:14:40.320)

From Germany to Japan to, it was actually, it did work.

Lex Fridman (1:14:44.480)

And so I appreciated the fact that PayPal existed

Lex Fridman (1:14:47.040)

and I had a way to get the money, the distribution was simple.

Lex Fridman (1:14:51.200)

This is pre Amazon book stuff.

Lex Fridman (1:14:53.480)

So it was just publishing a website.

Lex Fridman (1:14:55.320)

It was the popularity of SciPy emerging

Lex Fridman (1:14:57.120)

and getting company usage.

Lex Fridman (1:14:58.960)

I ended up not letting it go the five years

Lex Fridman (1:15:00.600)

and not trying to make the full amount

Lex Fridman (1:15:01.960)

because, you know, a year and a half later,

Travis Oliphant (1:15:04.560)

I was at Enthought.

Lex Fridman (1:15:05.400)

I had left academia as an Enthought

Lex Fridman (1:15:06.680)

and I kind of had a full time job.

Lex Fridman (1:15:07.880)

And then actually what happened is the documentation people,

Travis Oliphant (1:15:10.000)

there's a group that said, hey,

Lex Fridman (1:15:10.840)

we want to do documentation for SciPy as a collective.

Lex Fridman (1:15:14.280)

And they're essentially needing the stuff in the book, right?

Lex Fridman (1:15:18.680)

And so they kind of ask,

Lex Fridman (1:15:20.360)

hey, could we just use the stuff in your book?

Lex Fridman (1:15:21.920)

And at that point I said, yeah, I'll just open it up.

Lex Fridman (1:15:24.160)

So that's, but it has served its purpose.

Lex Fridman (1:15:27.320)

And the money that I made actually funded my grad student.

Travis Oliphant (1:15:31.040)

Like it was actually, you know,

Lex Fridman (1:15:32.160)

I paid him 25,000 a year out of that money.

Lex Fridman (1:15:35.440)

So the funny thing is if you do a very similar

Lex Fridman (1:15:37.440)

kind of experiment now with NumPy or something like it,

Travis Oliphant (1:15:40.680)

you could probably make a lot more.

Lex Fridman (1:15:42.480)

It's probably true.

Travis Oliphant (1:15:43.800)

Because of the tooling and the community building.

Lex Fridman (1:15:46.360)

Yeah, I agree.

Travis Oliphant (1:15:47.200)

Like the, and social media,

Lex Fridman (1:15:48.680)

that there's just a virality to that kind of idea.

Travis Oliphant (1:15:51.560)

I agree.

Lex Fridman (1:15:52.400)

There'd be things to do.

Travis Oliphant (1:15:53.240)

I've thought about that.

Lex Fridman (1:15:54.080)

And really I thought about a couple of books

Travis Oliphant (1:15:56.080)

or a couple of things that could be done there.

Lex Fridman (1:15:57.440)

And I just haven't, right?

Travis Oliphant (1:15:58.960)

Even, I tried to hire a ghostwriter this year too

Lex Fridman (1:16:01.920)

to see if that could help, but it didn't.

Lex Fridman (1:16:04.160)

But part of my problem is this,

Lex Fridman (1:16:06.240)

I've been so excited by a number of things

Travis Oliphant (1:16:08.080)

that have stemmed from that.

Lex Fridman (1:16:09.480)

Like, so I came here, worked at Enthought for four years,

Travis Oliphant (1:16:13.040)

graciously, Eric made me president.

Lex Fridman (1:16:14.960)

Then we started to work closely together.

Travis Oliphant (1:16:16.280)

We actually helped him buy out his partner.

Lex Fridman (1:16:19.440)

It didn't end great.

Travis Oliphant (1:16:20.720)

Like unfortunately Eric and I aren't real,

Lex Fridman (1:16:22.880)

aren't friends now.

Travis Oliphant (1:16:24.560)

I still respect him.

Lex Fridman (1:16:25.400)

I have a lot, I wish we were,

Lex Fridman (1:16:26.640)

but he didn't like the fact that Peter and I

Lex Fridman (1:16:30.240)

started Anaconda, right?

Travis Oliphant (1:16:31.680)

That was not, I mean, so there's two sides to that story.

Lex Fridman (1:16:36.200)

So I'm not gonna go into it, right?

Travis Oliphant (1:16:37.360)

Sure.

Lex Fridman (1:16:38.200)

But you, as human beings

Lex Fridman (1:16:40.600)

and you wish you still could be friends.

Lex Fridman (1:16:42.320)

I do, I do.

Travis Oliphant (1:16:43.920)

It saddens me.

Lex Fridman (1:16:45.160)

I mean, that's a story of great minds

Travis Oliphant (1:16:49.040)

building great companies.

Lex Fridman (1:16:51.480)

Somehow it's sad that when there's that kind of.

Lex Fridman (1:16:55.000)

And I hold him in esteem.

Lex Fridman (1:16:57.360)

I'm grateful for him.

Travis Oliphant (1:16:58.200)

I think Enthought still exists.

Lex Fridman (1:17:00.320)

They're doing great work helping scientists.

Travis Oliphant (1:17:02.520)

They still run the SciPy conference.

Lex Fridman (1:17:05.040)

They have an R&D platform they're selling now

Lex Fridman (1:17:07.320)

that's a tool that you can go get today, right?

Lex Fridman (1:17:10.080)

So Enthought has played a role in the SciPy

Travis Oliphant (1:17:14.920)

in supporting the community around SciPy, I would say.

Lex Fridman (1:17:18.240)

They ended up not being able to,

Travis Oliphant (1:17:20.560)

they ended up building a tool suite

Lex Fridman (1:17:22.040)

to write GUI applications.

Travis Oliphant (1:17:24.040)

Like that's where they could actually make

Lex Fridman (1:17:25.440)

that the business could work.

Lex Fridman (1:17:26.680)

And so supporting SciPy and NumPy itself

Lex Fridman (1:17:29.480)

wasn't as possible.

Travis Oliphant (1:17:30.560)

Like they didn't, they tried.

Lex Fridman (1:17:31.960)

I mean, it was not just because,

Travis Oliphant (1:17:33.280)

it was just because of the business aspect.

Lex Fridman (1:17:34.480)

So, and I wanted to build a company that could do,

Lex Fridman (1:17:36.840)

that could get venture funding, right?

Lex Fridman (1:17:39.080)

Better for worse.

Travis Oliphant (1:17:39.920)

I mean, that's a longer story.

Lex Fridman (1:17:41.040)

We could talk a lot about that, but.

Lex Fridman (1:17:42.400)

And that's where Anaconda came to be.

Lex Fridman (1:17:44.200)

That's where Anaconda came to be.

Lex Fridman (1:17:45.040)

So let me ask you, it's a little bit for fun

Lex Fridman (1:17:48.040)

because you built this amazing thing.

Lex Fridman (1:17:50.000)

And so let's talk about like an old warrior

Lex Fridman (1:17:54.640)

looking over old battles.

Travis Oliphant (1:17:57.320)

You've, you know, there's a sad letter in 2012

Lex Fridman (1:18:01.480)

that you wrote to the NumPy mailing list

Travis Oliphant (1:18:04.360)

announcing that you're leaving NumPy.

Lex Fridman (1:18:06.320)

And some of the things you've listed

Travis Oliphant (1:18:08.560)

as some of the things you regret

Lex Fridman (1:18:10.720)

or not regret necessarily, but some things to think about.

Travis Oliphant (1:18:14.440)

If you could go back and you could fix stuff about NumPy

Lex Fridman (1:18:17.640)

or both sort of in a personal level,

Lex Fridman (1:18:20.640)

but also like looking forward,

Lex Fridman (1:18:21.960)

what kind of things would you like to see changed?

Travis Oliphant (1:18:24.560)

Good question.

Lex Fridman (1:18:25.400)

So I think there's technical questions

Lex Fridman (1:18:26.320)

and social questions right there.

Lex Fridman (1:18:29.680)

First of all, you know, I wrote NumPy as a service

Lex Fridman (1:18:33.400)

and I spent a lot of time doing it.

Lex Fridman (1:18:35.000)

And then other people came help make it happen.

Lex Fridman (1:18:36.760)

NumPy succeeded because the work of a lot of people, right?

Lex Fridman (1:18:39.840)

So it's important to understand that.

Travis Oliphant (1:18:42.240)

I'm grateful for the opportunity,

Lex Fridman (1:18:43.880)

the role I had, I could play

Lex Fridman (1:18:45.080)

and grateful that things I did had an impact,

Lex Fridman (1:18:47.600)

but they only had the impact they had

Travis Oliphant (1:18:49.200)

because the other people that came to the story.

Lex Fridman (1:18:52.200)

And so they were essential,

Lex Fridman (1:18:53.440)

but the way data types were handled,

Lex Fridman (1:18:55.720)

the way data types, we had array scalers, for example,

Lex Fridman (1:18:59.280)

that are really just a substitute for a type concept, right?

Lex Fridman (1:19:04.080)

So we had array scalers or actual Python objects

Lex Fridman (1:19:06.960)

so that there's for every, for a 32 bit float

Lex Fridman (1:19:09.520)

or a 16 bit float or a 16 bit integer,

Travis Oliphant (1:19:13.160)

Python doesn't have a natural,

Lex Fridman (1:19:14.720)

it's just one integer, there's one float.

Travis Oliphant (1:19:17.040)

Well, what about these lower precision types,

Lex Fridman (1:19:19.960)

these larger precision types?

Lex Fridman (1:19:21.600)

So we had them in NumPy

Lex Fridman (1:19:23.680)

so that you could have a collection of them,

Lex Fridman (1:19:25.320)

but then have an object in Python that was one of them.

Lex Fridman (1:19:28.760)

And there's questions about like in retrospect,

Travis Oliphant (1:19:31.440)

I wouldn't have created those

Lex Fridman (1:19:32.920)

if it improved the type system.

Lex Fridman (1:19:34.880)

And like made the type system actually a Python type system

Lex Fridman (1:19:38.000)

as opposed to currently,

Travis Oliphant (1:19:39.480)

it's a Python one level type system.

Lex Fridman (1:19:41.400)

I don't know if you know the difference

Travis Oliphant (1:19:42.240)

between Python one, Python two,

Lex Fridman (1:19:43.200)

it's kind of technical, kind of depth,

Lex Fridman (1:19:44.880)

but Python two, one of its big things that Guido did,

Lex Fridman (1:19:47.320)

it was really brilliant.

Travis Oliphant (1:19:48.160)

It was the actually Python one,

Lex Fridman (1:19:51.640)

all classes, new objects were one.

Travis Oliphant (1:19:55.040)

If you as a user wrote a class,

Lex Fridman (1:19:56.880)

it was an instance of a single Python type

Lex Fridman (1:19:59.600)

called the class type, right?

Lex Fridman (1:20:02.000)

In Python two, he used a meta typing hook

Travis Oliphant (1:20:06.240)

to actually go, oh, we can extend this

Lex Fridman (1:20:07.960)

and have users write classes that are new types.

Lex Fridman (1:20:10.960)

So he was able to have your user classes be actual types

Lex Fridman (1:20:13.320)

and the Python type system got a lot more rich.

Travis Oliphant (1:20:16.480)

I barely understood that at the time that NumPy was written.

Lex Fridman (1:20:19.160)

And so I essentially in NumPy created a type system

Travis Oliphant (1:20:22.480)

that was Python one era.

Lex Fridman (1:20:24.400)

It was every D type is an instance of the same type

Travis Oliphant (1:20:29.240)

as opposed to having new D types be really just Python types

Lex Fridman (1:20:33.160)

with additional metadata.

Lex Fridman (1:20:34.280)

What's the cost of that?

Lex Fridman (1:20:35.440)

Is it efficiency, is it usability?

Travis Oliphant (1:20:37.200)

It's usability primarily.

Lex Fridman (1:20:38.840)

The cost isn't really efficiency.

Travis Oliphant (1:20:40.320)

It's the fact that it's clumsy to create new types.

Lex Fridman (1:20:45.080)

It's hard.

Lex Fridman (1:20:45.920)

And then one of the challenges,

Lex Fridman (1:20:47.560)

you wanna create new types.

Travis Oliphant (1:20:48.400)

You wanna quaternion type or you wanna add a new posit type

Lex Fridman (1:20:52.600)

or you wanna, so it's hard.

Lex Fridman (1:20:55.080)

And now, if we had done that well,

Lex Fridman (1:20:59.200)

when Numba came on the scene

Travis Oliphant (1:21:00.440)

where we could actually compile Python code,

Lex Fridman (1:21:02.880)

it would integrate with that type system much cleaner.

Lex Fridman (1:21:05.160)

And now all of a sudden you could do gradual typing

Lex Fridman (1:21:08.080)

more easily.

Travis Oliphant (1:21:08.920)

You could actually have Python when you add Numba

Lex Fridman (1:21:10.560)

plus better typing, could actually be a,

Travis Oliphant (1:21:14.720)

you'd smooth out a lot of rough edges.

Lex Fridman (1:21:16.800)

But there's already, there's like,

Lex Fridman (1:21:18.840)

but are you talking about from the perspective

Lex Fridman (1:21:20.960)

of developers within NumPy or users of NumPy?

Travis Oliphant (1:21:23.840)

Developers of new, not really users of NumPy so much.

Lex Fridman (1:21:27.080)

It's the development of NumPy.

Lex Fridman (1:21:28.800)

So you're thinking about like how to design NumPy

Lex Fridman (1:21:32.160)

so that it's contributors.

Travis Oliphant (1:21:33.880)

Yeah, the contributors, it's easier.

Lex Fridman (1:21:35.880)

It's easier.

Travis Oliphant (1:21:36.720)

It's less work to make it better and to keep it maintained.

Lex Fridman (1:21:39.320)

And where that's impacted things, for example,

Travis Oliphant (1:21:41.480)

is the GPU.

Lex Fridman (1:21:43.400)

Like all of a sudden GPUs start getting added

Lex Fridman (1:21:45.520)

and we don't have them in NumPy.

Lex Fridman (1:21:48.360)

Like NumPy should just work on GPUs.

Travis Oliphant (1:21:50.560)

The fact that we'd have to download a whole other object

Lex Fridman (1:21:52.680)

called Kupy to have arrays on GPUs

Travis Oliphant (1:21:54.800)

is just an artifact of history.

Lex Fridman (1:21:57.440)

Like there's no fundamental reason for it.

Travis Oliphant (1:21:59.160)

Well, that's really interesting.

Lex Fridman (1:22:00.200)

If we could sort of go on that tangent briefly

Travis Oliphant (1:22:02.520)

is you have PyTorch and other libraries like TensorFlow

Lex Fridman (1:22:07.800)

that basically tried to mimic NumPy.

Travis Oliphant (1:22:11.840)

Like you've created a sort of platonic form

Lex Fridman (1:22:15.720)

of multi dimension. Basically, yeah.

Travis Oliphant (1:22:16.920)

Yeah, exactly.

Lex Fridman (1:22:17.760)

Well, and the problem was I didn't realize that.

Travis Oliphant (1:22:19.800)

Platonic form has a lot of edges.

Lex Fridman (1:22:21.760)

They're like, well, we should cut those out

Travis Oliphant (1:22:23.360)

before we present it.

Lex Fridman (1:22:24.200)

So I wonder if you can comment,

Lex Fridman (1:22:26.920)

is there like a difference between their implementations?

Lex Fridman (1:22:29.360)

Do you wish that they were all using NumPy

Lex Fridman (1:22:31.440)

or like in this abstraction of GPU?

Lex Fridman (1:22:34.040)

And sorry to interrupt that there's GPUs, ASICs.

Travis Oliphant (1:22:38.240)

There might be other neuromorphic computing.

Lex Fridman (1:22:40.040)

There might be other kind of,

Travis Oliphant (1:22:41.600)

or the aliens will come with a new kind of computer.

Lex Fridman (1:22:43.920)

Like an abstraction that NumPy should just operate nicely

Travis Oliphant (1:22:47.880)

over the things that are more and more

Lex Fridman (1:22:50.280)

and smarter and smarter with this multi dimensional arrays.

Travis Oliphant (1:22:54.200)

Yeah, yeah.

Lex Fridman (1:22:55.520)

There's several comments there.

Travis Oliphant (1:22:56.920)

We are working on something now called data dash APIs.org.

Lex Fridman (1:23:00.360)

Data dash API.org, you can go there today.

Lex Fridman (1:23:02.560)

And it's our answer.

Lex Fridman (1:23:04.480)

It's my answer.

Travis Oliphant (1:23:05.320)

It's not just me.

Lex Fridman (1:23:06.160)

It's me and Rolf and Athen and Aaron

Lex Fridman (1:23:09.120)

and a lot of companies are helping us at Quansight Labs.

Lex Fridman (1:23:13.120)

It's not unifying all the arrays.

Travis Oliphant (1:23:14.560)

It's creating an API that is unified.

Lex Fridman (1:23:17.200)

So we do care about this

Lex Fridman (1:23:19.360)

and we're trying to work through it.

Lex Fridman (1:23:21.280)

I actually had the chance to go and meet

Travis Oliphant (1:23:22.560)

with the TensorFlow team and the PyTorch team

Lex Fridman (1:23:25.360)

and talk to them after exiting Anaconda.

Travis Oliphant (1:23:29.120)

Just talking about,

Lex Fridman (1:23:29.960)

because the first year after leaving Anaconda in 2018,

Travis Oliphant (1:23:33.960)

I became deeply aware of this and realized that,

Lex Fridman (1:23:36.000)

oh, this split in the array community that exists today

Travis Oliphant (1:23:38.960)

makes what I was concerned about in 2005 pretty parochial.

Lex Fridman (1:23:44.160)

It's a lot worse, right?

Travis Oliphant (1:23:45.880)

Now there's a lot more people.

Lex Fridman (1:23:47.280)

So perhaps the industry can sustain more stacks, right?

Travis Oliphant (1:23:51.400)

There's a lot of money,

Lex Fridman (1:23:52.560)

but it makes it a lot less efficient.

Travis Oliphant (1:23:54.120)

I mean, but I've also learned to appreciate,

Lex Fridman (1:23:56.720)

it's okay to have some competition.

Travis Oliphant (1:23:58.440)

It's okay to have different implementations,

Lex Fridman (1:24:00.760)

but it's better if you can at least refactor some parts.

Travis Oliphant (1:24:03.560)

I mean, you're gonna be more efficient

Lex Fridman (1:24:04.960)

if you can refactor parts.

Travis Oliphant (1:24:07.000)

It's nice to have competition over things,

Lex Fridman (1:24:09.560)

over what is nice to have competition.

Travis Oliphant (1:24:11.760)

They're innovative.

Lex Fridman (1:24:12.600)

Yeah, innovative.

Lex Fridman (1:24:13.440)

And then maybe on the infrastructure,

Lex Fridman (1:24:15.920)

whatever, however you define infrastructure,

Travis Oliphant (1:24:18.120)

that maybe it's nice to have come together.

Lex Fridman (1:24:21.400)

Exactly, I agree.

Lex Fridman (1:24:22.440)

And I think, but it was interesting to hear the stories.

Lex Fridman (1:24:24.600)

I mean, TensorFlow came out of a C++ library,

Travis Oliphant (1:24:29.040)

Jeff Dean wrote, I think,

Lex Fridman (1:24:30.160)

that was basically how they were doing inference, right?

Lex Fridman (1:24:33.560)

And then they realized, oh,

Lex Fridman (1:24:34.400)

we could do this TensorFlow thing.

Travis Oliphant (1:24:36.440)

That C++ library, then what was interesting to me

Lex Fridman (1:24:38.400)

was the fact that both Google and Facebook did not,

Travis Oliphant (1:24:42.600)

it's not like they supported Python or NumPy initially.

Lex Fridman (1:24:44.960)

They just realized they had to.

Travis Oliphant (1:24:47.200)

They came to this world and then all the users were like,

Lex Fridman (1:24:48.760)

hey, where's the NumPy interface?

Travis Oliphant (1:24:50.680)

Oh, and then they kind of came late to it

Lex Fridman (1:24:52.560)

and then they had these bolt ons.

Travis Oliphant (1:24:54.800)

TensorFlow's bolt on, I don't mean to offend,

Lex Fridman (1:24:57.280)

but it was so bad.

Travis Oliphant (1:24:58.480)

Yeah, it was bad.

Lex Fridman (1:24:59.320)

It's the first time that I'm usually,

Travis Oliphant (1:25:01.760)

I mean, one of the challenges I have

Lex Fridman (1:25:04.160)

is I don't criticize enough in the sense

Travis Oliphant (1:25:07.000)

that I don't give people input enough, you know, if.

Lex Fridman (1:25:09.960)

I think it's universally agreed upon

Travis Oliphant (1:25:11.680)

that the bolt ons on TensorFlow were.

Lex Fridman (1:25:13.640)

But I went to, it was a talk given at Mallorca in Spain

Lex Fridman (1:25:17.080)

and a great guy came and gave a talk and I said,

Lex Fridman (1:25:19.880)

you should never show that API again

Travis Oliphant (1:25:21.400)

at a PyData conference.

Lex Fridman (1:25:23.040)

Like that was, that's terrible.

Travis Oliphant (1:25:24.840)

Like you're taking this beautiful system we've created

Lex Fridman (1:25:27.080)

and like you're corrupting all these poor Python people,

Travis Oliphant (1:25:29.440)

forcing them to write code like that

Lex Fridman (1:25:30.840)

or thinking they should.

Travis Oliphant (1:25:32.640)

Fortunately, you know, they adopted Keras as their,

Lex Fridman (1:25:35.640)

and Keras is better.

Lex Fridman (1:25:36.760)

And so Keras, TensorFlow is fine, is reasonable,

Lex Fridman (1:25:40.360)

but they bolted it on.

Travis Oliphant (1:25:42.680)

Facebook did too.

Lex Fridman (1:25:43.640)

Like Facebook had their own C++ library for doing inference

Lex Fridman (1:25:48.160)

and they also had the same reaction, they had to do this.

Lex Fridman (1:25:51.160)

One big difference is Facebook,

Travis Oliphant (1:25:52.840)

maybe because of the way it's situated in part of fair,

Lex Fridman (1:25:55.240)

part of the research library,

Travis Oliphant (1:25:56.600)

TensorFlow is definitely used and, you know,

Lex Fridman (1:25:58.880)

they have to make, they couldn't just open it up

Lex Fridman (1:26:00.720)

and let the community, you know, change what that is.

Lex Fridman (1:26:03.160)

Cause I guess they were worried

Travis Oliphant (1:26:04.640)

about disrupting their operations.

Lex Fridman (1:26:06.880)

Facebook's been much more open to having community input

Travis Oliphant (1:26:10.720)

on the structure itself.

Lex Fridman (1:26:12.400)

Whereas Google and TensorFlow,

Travis Oliphant (1:26:14.240)

they're really eager to have community users,

Lex Fridman (1:26:16.000)

people use it and build the infrastructure,

Lex Fridman (1:26:17.520)

but it's much more walled.

Lex Fridman (1:26:18.840)

Like it's harder to become a contributor to TensorFlow.

Lex Fridman (1:26:21.600)

And it's also, this is very difficult question to answer

Lex Fridman (1:26:24.760)

and don't mean to be throwing shade at anybody,

Lex Fridman (1:26:27.080)

but you have to wonder, it's the Microsoft question

Lex Fridman (1:26:30.320)

of when you have a tool like PyTorch or TensorFlow,

Lex Fridman (1:26:33.920)

how much are you tending to the hackers

Lex Fridman (1:26:36.320)

and how much are you tending to the big corporate clients?

Travis Oliphant (1:26:39.240)

Correct.

Lex Fridman (1:26:40.080)

So like the ones that,

Lex Fridman (1:26:42.560)

do you tend to the millions of people

Lex Fridman (1:26:44.160)

that are giving you almost no money,

Travis Oliphant (1:26:46.440)

or do you tend to the few

Lex Fridman (1:26:48.360)

that are giving you a ton of money?

Travis Oliphant (1:26:50.320)

I tend to stand with the people.

Lex Fridman (1:26:54.000)

Right.

Travis Oliphant (1:26:54.840)

Cause I feel like if you nurture the hackers,

Lex Fridman (1:26:57.760)

you will make the right decisions in the longterm

Travis Oliphant (1:27:00.200)

that will make the companies happy.

Lex Fridman (1:27:02.000)

I lean that way too.

Travis Oliphant (1:27:03.280)

I totally agree.

Lex Fridman (1:27:04.120)

But then you have to find the right dance.

Lex Fridman (1:27:05.680)

But it's a balance.

Lex Fridman (1:27:07.080)

Cause you can lean to the hackers and run out of money.

Travis Oliphant (1:27:08.960)

Yeah, exactly.

Lex Fridman (1:27:10.240)

Exactly.

Travis Oliphant (1:27:11.440)

Which has been some of the challenge I've faced

Lex Fridman (1:27:13.760)

in the sense that,

Travis Oliphant (1:27:14.680)

like I would look at some of the experiments,

Lex Fridman (1:27:17.040)

like NumPy, the fact that we have this split

Travis Oliphant (1:27:19.040)

is a factor of I wasn't able to collect more money

Lex Fridman (1:27:21.720)

towards NumPy development.

Travis Oliphant (1:27:22.800)

Yeah.

Lex Fridman (1:27:23.640)

Right?

Travis Oliphant (1:27:24.480)

I mean, I didn't succeed in the early days

Lex Fridman (1:27:26.480)

of getting enough financial contribution to NumPy

Lex Fridman (1:27:29.560)

so that they could work on it.

Lex Fridman (1:27:31.080)

Right?

Travis Oliphant (1:27:31.920)

I couldn't work on it full time.

Lex Fridman (1:27:32.760)

I had to just catch an hour here, an hour there.

Lex Fridman (1:27:35.640)

And I basically not liked that.

Lex Fridman (1:27:37.880)

Like I've wanted to be able to do something about that

Travis Oliphant (1:27:39.920)

for a long time and try to figure out how,

Lex Fridman (1:27:41.440)

well, there's lots of ways.

Travis Oliphant (1:27:42.960)

I mean, possibly one could say,

Lex Fridman (1:27:44.640)

we had an offer from Microsoft

Travis Oliphant (1:27:46.240)

at early days of Anaconda.

Lex Fridman (1:27:48.240)

2014, they offered to come buy us, right?

Travis Oliphant (1:27:51.160)

The problem was the right people at Microsoft

Lex Fridman (1:27:52.760)

didn't offer to buy us.

Lex Fridman (1:27:53.600)

And they were still,

Lex Fridman (1:27:54.880)

they were, it was really a,

Travis Oliphant (1:27:56.440)

we were like a second,

Lex Fridman (1:27:58.040)

they had really bought, they just bought R,

Travis Oliphant (1:27:59.680)

the R company called,

Lex Fridman (1:28:01.800)

it was not R studio,

Lex Fridman (1:28:02.800)

but it was another R company that was emergent.

Lex Fridman (1:28:05.680)

And it was kind of a,

Travis Oliphant (1:28:07.160)

well, we should also get a Python play,

Lex Fridman (1:28:09.360)

but they were really doubling down on R.

Lex Fridman (1:28:11.520)

Right?

Lex Fridman (1:28:12.360)

And so it was like,

Travis Oliphant (1:28:13.200)

it was where you would go to die.

Lex Fridman (1:28:14.400)

So it's not, it wasn't,

Travis Oliphant (1:28:15.440)

it was before Satya was there.

Lex Fridman (1:28:17.160)

Satya had just started.

Travis Oliphant (1:28:18.680)

Just started.

Lex Fridman (1:28:19.520)

Right?

Lex Fridman (1:28:20.360)

And the offer was coming from someone

Lex Fridman (1:28:21.800)

two levels down from him.

Travis Oliphant (1:28:23.080)

Got you.

Lex Fridman (1:28:23.920)

Right?

Lex Fridman (1:28:24.760)

And if it had come from Scott Guthrie,

Lex Fridman (1:28:26.640)

so I got a chance to meet Scott Guthrie,

Travis Oliphant (1:28:28.320)

great guy, I like him.

Lex Fridman (1:28:29.760)

If an offer had come from him,

Travis Oliphant (1:28:31.560)

probably would be at Microsoft right now.

Lex Fridman (1:28:33.200)

That'd be fascinating.

Travis Oliphant (1:28:34.520)

That would be really nice actually,

Lex Fridman (1:28:36.160)

especially given what Microsoft has since done

Travis Oliphant (1:28:38.720)

for the open source community and all those things.

Lex Fridman (1:28:40.200)

Yes, I think they're doing well.

Travis Oliphant (1:28:41.640)

I really like some of the stuff they've been doing.

Lex Fridman (1:28:43.720)

They're still working,

Lex Fridman (1:28:45.200)

and they've, you know,

Lex Fridman (1:28:46.040)

they've hired Guido now,

Lex Fridman (1:28:46.880)

and they've hired a lot of Python developers.

Lex Fridman (1:28:47.720)

Wait, Guido's not at Microsoft?

Travis Oliphant (1:28:49.400)

Yeah, he works at Microsoft.

Lex Fridman (1:28:50.240)

I need to.

Travis Oliphant (1:28:52.480)

Which, he retired,

Lex Fridman (1:28:53.600)

then he came out of retirement,

Lex Fridman (1:28:54.720)

and he's working now.

Lex Fridman (1:28:55.560)

I was just talking to him,

Lex Fridman (1:28:56.400)

and he didn't mention this person.

Lex Fridman (1:28:57.840)

Well.

Travis Oliphant (1:28:58.680)

I should investigate this further.

Lex Fridman (1:29:01.280)

Well.

Travis Oliphant (1:29:02.120)

Because I know he loved Dropbox,

Lex Fridman (1:29:02.960)

but I wasn't sure what he was doing,

Travis Oliphant (1:29:04.000)

who he was up to.

Lex Fridman (1:29:05.160)

Well, he was kind of saying he'd retire,

Travis Oliphant (1:29:06.560)

but, and it's literally been five years

Lex Fridman (1:29:09.640)

since I last sat down and really talked to Guido.

Lex Fridman (1:29:12.280)

Right?

Lex Fridman (1:29:13.640)

Guido's a technology expert, right?

Travis Oliphant (1:29:16.000)

He's a, so I came,

Lex Fridman (1:29:17.480)

I was excited because I'd finally figured out

Travis Oliphant (1:29:18.880)

the type system for NumPy.

Lex Fridman (1:29:20.720)

I wanted to kind of talk about that with him,

Lex Fridman (1:29:22.240)

and I kind of overwhelmed him.

Lex Fridman (1:29:23.960)

Could you stay in that,

Travis Oliphant (1:29:25.080)

just for a brief moment,

Lex Fridman (1:29:26.640)

because you're a fascinating person

Travis Oliphant (1:29:28.200)

in the history of programming.

Lex Fridman (1:29:29.440)

He is a fascinating person.

Lex Fridman (1:29:31.240)

What have you learned from Guido

Lex Fridman (1:29:34.200)

about programming, about life?

Travis Oliphant (1:29:37.560)

Yeah, yeah.

Lex Fridman (1:29:38.400)

A lot, actually.

Travis Oliphant (1:29:39.240)

I've been a fan of Guido's.

Lex Fridman (1:29:40.840)

You know, we have a chance to talk.

Travis Oliphant (1:29:42.520)

Some, I wouldn't say, you know,

Lex Fridman (1:29:43.760)

we talk all the time.

Travis Oliphant (1:29:44.840)

Not at all.

Lex Fridman (1:29:45.680)

He may, but we talk enough to,

Travis Oliphant (1:29:47.520)

I respect his,

Lex Fridman (1:29:48.840)

in fact, when I first started NumPy,

Travis Oliphant (1:29:49.880)

one of the first things I did was I had a,

Lex Fridman (1:29:51.520)

I asked Guido for a meeting

Travis Oliphant (1:29:53.320)

with him and Paul Dubois in San Mateo.

Lex Fridman (1:29:55.400)

And I went and met him for lunch.

Lex Fridman (1:29:56.920)

And basically, to say,

Lex Fridman (1:29:58.000)

maybe we can actually,

Travis Oliphant (1:29:59.200)

part of the strategy for NumPy

Lex Fridman (1:30:00.720)

was to get it into Python 3,

Lex Fridman (1:30:02.440)

and maybe be part of Python.

Lex Fridman (1:30:04.120)

And so we talked about that.

Travis Oliphant (1:30:05.160)

That's a cool conversation.

Lex Fridman (1:30:06.000)

And about that approach, right?

Travis Oliphant (1:30:06.920)

I would have loved to be a flyer in the water.

Lex Fridman (1:30:09.200)

That was good.

Lex Fridman (1:30:10.040)

And over the years for Guido,

Lex Fridman (1:30:12.080)

I learned,

Lex Fridman (1:30:13.560)

so he was open.

Lex Fridman (1:30:14.840)

Like, he was willing to listen to people's ideas.

Lex Fridman (1:30:18.200)

Right?

Lex Fridman (1:30:19.040)

And over the years,

Travis Oliphant (1:30:19.880)

now generally, you know,

Lex Fridman (1:30:20.920)

I'm not saying universally that's been true,

Lex Fridman (1:30:22.600)

but generally that's been true.

Lex Fridman (1:30:24.360)

So he's willing to listen.

Travis Oliphant (1:30:25.680)

He's willing to defer.

Lex Fridman (1:30:27.240)

Like on the scientific side,

Travis Oliphant (1:30:28.280)

he would just kind of defer.

Lex Fridman (1:30:29.120)

He didn't really always understand

Lex Fridman (1:30:30.160)

what we were doing.

Lex Fridman (1:30:31.000)

Yeah.

Lex Fridman (1:30:31.840)

And he'd defer.

Lex Fridman (1:30:32.800)

One place where he didn't enough

Travis Oliphant (1:30:35.640)

was we missed a matrix multiply operator.

Lex Fridman (1:30:37.680)

Like that finally got added to Python,

Lex Fridman (1:30:39.600)

but about 10 years later than it should have.

Lex Fridman (1:30:42.240)

But the reason was because nobody,

Travis Oliphant (1:30:44.760)

it takes a lot of effort.

Lex Fridman (1:30:46.200)

And I learned this while I was writing NumPy.

Travis Oliphant (1:30:48.160)

I also wrote tools to Python.

Lex Fridman (1:30:49.320)

I began with Python Dev,

Lex Fridman (1:30:50.160)

and I added some pieces to Python.

Lex Fridman (1:30:52.320)

Like the memory view object.

Travis Oliphant (1:30:53.400)

I wanted the structure of NumPy into Python.

Lex Fridman (1:30:55.680)

So we didn't get NumPy into Python,

Lex Fridman (1:30:56.960)

but we got the basic structure of it into Python.

Lex Fridman (1:30:59.480)

Like, so you could build on it.

Travis Oliphant (1:31:01.000)

Nobody did for a while,

Lex Fridman (1:31:01.880)

but eventually database authors started to.

Lex Fridman (1:31:04.720)

And it's a lot better.

Lex Fridman (1:31:05.760)

They did.

Lex Fridman (1:31:06.600)

And also Antoine Petrou and Stefan Krah

Lex Fridman (1:31:08.960)

actually fixed the memory view object.

Travis Oliphant (1:31:10.760)

Cause I wrote the underlying infrastructure in C,

Lex Fridman (1:31:13.280)

but the Python exposure was terrible

Travis Oliphant (1:31:15.520)

until they came in and fixed it.

Lex Fridman (1:31:16.640)

Partly because I was writing NumPy,

Lex Fridman (1:31:18.080)

and NumPy was the Python exposure.

Lex Fridman (1:31:19.960)

I didn't really care about

Travis Oliphant (1:31:21.240)

if you didn't have NumPy installed.

Lex Fridman (1:31:22.800)

Anyway, Guido opened up ideas,

Travis Oliphant (1:31:25.360)

technologically brilliant.

Lex Fridman (1:31:27.280)

Like really, I really got a lot of respect for him

Travis Oliphant (1:31:29.440)

when I saw what he did

Lex Fridman (1:31:30.360)

with this type class merger thing.

Lex Fridman (1:31:33.320)

It was actually tricky, right?

Lex Fridman (1:31:35.200)

And then willing to share, willing to share his ideas.

Lex Fridman (1:31:38.400)

So the other thing early on in 1998,

Lex Fridman (1:31:40.200)

I said, I wrote my first extension module.

Travis Oliphant (1:31:42.240)

The reason I could is because he'd written this blog post

Lex Fridman (1:31:44.800)

on how to do reference counting, right?

Lex Fridman (1:31:47.360)

And without it, I would have been lost, right?

Lex Fridman (1:31:50.040)

But he was willing to at least try to write this post.

Lex Fridman (1:31:53.240)

And so he's been motivated early on with Python.

Lex Fridman (1:31:56.080)

There's a computer science for everybody.

Travis Oliphant (1:31:58.200)

You kind of have this early on desire to,

Lex Fridman (1:31:59.880)

oh, maybe we should be pushing programming to more people.

Lex Fridman (1:32:02.040)

So he had this populist notion, I guess,

Lex Fridman (1:32:04.560)

or populist sense to learn that there's a certain skill,

Lex Fridman (1:32:08.720)

and I've seen it in other people too,

Lex Fridman (1:32:10.560)

of engaging with contributors sufficiently to,

Travis Oliphant (1:32:13.960)

because when somebody engaged with you

Lex Fridman (1:32:15.640)

and wants to contribute to you,

Travis Oliphant (1:32:16.480)

if you ignore them, they go away.

Lex Fridman (1:32:18.400)

So building that early contributor base

Travis Oliphant (1:32:19.760)

requires real engagement with other people.

Lex Fridman (1:32:23.320)

And he would do that.

Lex Fridman (1:32:24.520)

Can you also comment on this tragic stepping down

Lex Fridman (1:32:29.080)

from his position as the benevolent dictator for life

Lex Fridman (1:32:32.880)

over the wars, you know?

Lex Fridman (1:32:35.640)

The Walrus operator?

Travis Oliphant (1:32:36.560)

The Walrus operator was the last battle.

Lex Fridman (1:32:39.200)

I don't know if that's the cause of it,

Lex Fridman (1:32:40.880)

but there's this, for people who don't know,

Lex Fridman (1:32:43.640)

you can look up, there's the Walrus operator,

Travis Oliphant (1:32:45.640)

which looks like a colon and equal sign.

Lex Fridman (1:32:49.560)

Yeah, colon, equal sign.

Lex Fridman (1:32:50.800)

And it actually does maybe the thing

Lex Fridman (1:32:54.680)

that an equal sign should be doing.

Travis Oliphant (1:32:57.560)

Yeah, maybe, right, exactly.

Lex Fridman (1:33:00.240)

But it's just historically,

Travis Oliphant (1:33:02.080)

equal sign means something else.

Lex Fridman (1:33:03.560)

It just means assignment.

Lex Fridman (1:33:05.240)

So he stepped down over this.

Lex Fridman (1:33:07.280)

What do you think about the pressure of leadership?

Travis Oliphant (1:33:10.360)

It's something that, you mentioned the letter I wrote

Lex Fridman (1:33:12.280)

in NumPy at the time.

Travis Oliphant (1:33:13.640)

That was a hard time, actually.

Lex Fridman (1:33:15.240)

I mean, there's been really hard times.

Travis Oliphant (1:33:17.080)

It was hard.

Lex Fridman (1:33:19.520)

You get criticized, right?

Lex Fridman (1:33:20.840)

And you get pushed, and you get,

Lex Fridman (1:33:22.800)

not everybody loves what you do.

Travis Oliphant (1:33:23.800)

Like anytime you do anything that has impact at all,

Lex Fridman (1:33:26.880)

you're not universally loved, right?

Travis Oliphant (1:33:28.560)

You get some real critics.

Lex Fridman (1:33:29.760)

And that's an important energy,

Travis Oliphant (1:33:31.960)

because it's impossible for you to do everything right.

Lex Fridman (1:33:35.080)

You need people to be pushing.

Lex Fridman (1:33:37.160)

But sometimes people can get mean, right?

Lex Fridman (1:33:39.320)

People can, I prefer to give people the benefit of the doubt.

Travis Oliphant (1:33:43.080)

I don't immediately assume they have bad intentions.

Lex Fridman (1:33:45.800)

And maybe for other, maybe that doesn't happen for everybody.

Travis Oliphant (1:33:49.000)

For whatever reason, their past,

Lex Fridman (1:33:50.200)

their experiences with people, they sometimes have bad,

Lex Fridman (1:33:53.040)

so they immediately attribute to you bad intentions.

Lex Fridman (1:33:54.880)

So you're like, where did this come from?

Travis Oliphant (1:33:56.080)

I mean, I'm definitely open to criticism,

Lex Fridman (1:33:57.760)

but I think you're misinterpreting the whole point.

Travis Oliphant (1:34:00.520)

Because I would get that, certainly when I started Anaconda.

Lex Fridman (1:34:05.800)

Sometimes I say to people,

Travis Oliphant (1:34:08.520)

I care enough about entrepreneurship

Lex Fridman (1:34:09.760)

to make some open source people uncomfortable.

Lex Fridman (1:34:12.240)

And I care enough about open source

Lex Fridman (1:34:13.520)

to make investors uncomfortable.

Lex Fridman (1:34:15.560)

So I sort of, you create kind of doubters on both sides.

Lex Fridman (1:34:19.880)

So when you have, and this is just a plea

Travis Oliphant (1:34:23.840)

to the listener and the public, I've noticed this too,

Lex Fridman (1:34:27.680)

that there's a tendency, and social media makes this worse,

Travis Oliphant (1:34:32.680)

when you don't have perfect information about the situation,

Lex Fridman (1:34:35.560)

you tend to fill the gaps with the worst possible,

Travis Oliphant (1:34:39.280)

or at least a bad story that fills those gaps.

Lex Fridman (1:34:43.080)

And I think it's good to live life,

Travis Oliphant (1:34:46.960)

maybe not fully naively, but filling in the gaps

Lex Fridman (1:34:49.760)

with the good, with the best, with the positive,

Travis Oliphant (1:34:54.720)

with the hopeful explanation of why you see this.

Lex Fridman (1:34:57.280)

So if you see somebody like you trying to make money

Travis Oliphant (1:35:00.280)

on a book about an umpire,

Lex Fridman (1:35:01.960)

there's a million stories around that that are positive.

Lex Fridman (1:35:04.880)

And those are good to think about,

Lex Fridman (1:35:07.840)

to project positive intent on the people.

Travis Oliphant (1:35:10.600)

Because for many reasons, usually because people are good

Lex Fridman (1:35:13.960)

and they do have good intent.

Lex Fridman (1:35:15.560)

And also when you project that positive intent,

Lex Fridman (1:35:17.480)

people will step up to that too.

Travis Oliphant (1:35:19.400)

Yes.

Lex Fridman (1:35:20.240)

It's a great point.

Travis Oliphant (1:35:21.760)

It has this kind of viral nature to it.

Lex Fridman (1:35:24.320)

And of course with Twitter, early on figured out,

Lex Fridman (1:35:27.720)

and Facebook is that they can make a lot of money

Lex Fridman (1:35:30.360)

and engagement from the negative.

Travis Oliphant (1:35:32.280)

Yes.

Lex Fridman (1:35:33.120)

So there's this, we're fighting this mechanism.

Travis Oliphant (1:35:35.440)

I agree.

Lex Fridman (1:35:36.280)

Which is challenging.

Travis Oliphant (1:35:37.120)

It's easier.

Lex Fridman (1:35:37.940)

It's just easier to be.

Travis Oliphant (1:35:38.780)

To be negative.

Lex Fridman (1:35:39.620)

And then for some reason, something in our minds

Travis Oliphant (1:35:41.920)

really enjoys sharing that and getting all excited

Lex Fridman (1:35:45.280)

about the negativity.

Travis Oliphant (1:35:46.280)

We do, yeah.

Lex Fridman (1:35:47.400)

Some protective mechanism perhaps that we're gonna get eaten

Travis Oliphant (1:35:50.440)

if we don't, yeah.

Lex Fridman (1:35:51.280)

Exactly.

Travis Oliphant (1:35:52.100)

For us to be effective as a group of people

Lex Fridman (1:35:53.200)

in a software engineering project,

Travis Oliphant (1:35:54.600)

you have to project positive intent, I think.

Lex Fridman (1:35:56.860)

I totally agree.

Travis Oliphant (1:35:57.820)

Totally agree.

Lex Fridman (1:35:58.660)

And I think that's very,

Lex Fridman (1:35:59.480)

and so that happens in this space.

Lex Fridman (1:36:01.640)

But Python has done a reasonable job in the past,

Lex Fridman (1:36:03.840)

but here is a situation where I think it started

Lex Fridman (1:36:05.920)

to get this pressure where it didn't.

Travis Oliphant (1:36:07.840)

I really didn't, I didn't know enough about what happened.

Lex Fridman (1:36:10.440)

I've talked to several people about it.

Lex Fridman (1:36:12.160)

And I know most of the steering committee members today,

Lex Fridman (1:36:15.840)

one person nominated me for that role,

Lex Fridman (1:36:17.880)

but it's the wrong role for me right now, right?

Lex Fridman (1:36:20.880)

I have a lot of respect for the Python developer space

Lex Fridman (1:36:24.040)

and the Python developers.

Lex Fridman (1:36:25.440)

I also understand the gap between computer science

Travis Oliphant (1:36:27.600)

Python developers and array programming developers

Lex Fridman (1:36:30.440)

or science developers.

Lex Fridman (1:36:31.440)

And in fact, Python succeeds in the array space

Lex Fridman (1:36:34.560)

the more it has people in that boundary.

Lex Fridman (1:36:36.520)

And there's often very few.

Lex Fridman (1:36:37.960)

Like I was playing a role in that boundary

Lex Fridman (1:36:39.440)

and working like everything to try to keep up

Lex Fridman (1:36:42.600)

with even what Guido was saying, like I'm a C programmer,

Lex Fridman (1:36:47.720)

but not a computer scientist.

Lex Fridman (1:36:49.080)

Like I was an engineer and physicist and mathematician,

Lex Fridman (1:36:52.600)

and I didn't always understand

Lex Fridman (1:36:54.840)

what they were talking about

Lex Fridman (1:36:56.360)

and why they would have opinions the way they did.

Lex Fridman (1:36:58.360)

So, you know, you have to listen and try to understand.

Travis Oliphant (1:37:00.280)

Then you also have to explain your point of view

Lex Fridman (1:37:02.120)

in a way they can understand.

Lex Fridman (1:37:03.560)

And that takes a lot of work.

Lex Fridman (1:37:04.840)

And that communication is always the challenge.

Lex Fridman (1:37:07.920)

And it's just what we're describing here

Lex Fridman (1:37:09.200)

about the negativity is just another form of that.

Lex Fridman (1:37:11.520)

Like how do we come together?

Lex Fridman (1:37:12.560)

And it does appear we're wired anyway

Travis Oliphant (1:37:14.520)

to at least have a, there's a part of us

Lex Fridman (1:37:16.560)

that will enemy, you know, friend, enemy.

Lex Fridman (1:37:18.880)

And we see, yeah, it's like,

Lex Fridman (1:37:21.360)

why are we wiring on the enemy front?

Lex Fridman (1:37:23.520)

So why are we pushing that?

Lex Fridman (1:37:24.760)

Why are we promoting that so deeply?

Travis Oliphant (1:37:26.680)

Assume friend until proven otherwise.

Lex Fridman (1:37:28.440)

Yes, yes.

Travis Oliphant (1:37:30.000)

So, cause you have such a fascinating mind in all of this.

Lex Fridman (1:37:32.160)

Let me just ask you these questions.

Lex Fridman (1:37:34.160)

So one interesting side on the Python history

Lex Fridman (1:37:38.000)

is the move from Python two to Python three.

Travis Oliphant (1:37:41.000)

You mentioned move from Python one to Python two,

Lex Fridman (1:37:43.720)

but the move from Python two to Python three

Travis Oliphant (1:37:46.800)

is a little bit interesting

Lex Fridman (1:37:47.920)

because it took a very long time.

Travis Oliphant (1:37:50.040)

It broke, you know, quite a small way

Lex Fridman (1:37:53.520)

backward compatibility, but even that small way

Travis Oliphant (1:37:56.280)

seemed to have been very painful for people.

Lex Fridman (1:37:58.680)

Is there lessons you draw?

Travis Oliphant (1:38:00.640)

Oh man, tons of lessons.

Lex Fridman (1:38:01.480)

From how long it took and how painful it seemed to be?

Travis Oliphant (1:38:05.520)

Yeah, tons of lessons.

Lex Fridman (1:38:07.000)

Well, I mentioned here earlier

Travis Oliphant (1:38:08.240)

that NumPy was written in 2005.

Lex Fridman (1:38:11.840)

It was in 2005 that I actually went to Guido

Travis Oliphant (1:38:15.520)

to talk about getting NumPy into Python three.

Lex Fridman (1:38:17.240)

Like my strategy was to,

Travis Oliphant (1:38:18.880)

oh, we were moving to Python three.

Lex Fridman (1:38:19.960)

Let's have that be, and it seems funny in retrospect

Travis Oliphant (1:38:22.200)

because like, wait, Python three,

Lex Fridman (1:38:23.360)

that was in 2020, right?

Travis Oliphant (1:38:25.480)

When we finally ended the support for Python two

Lex Fridman (1:38:27.760)

or at least 2017.

Travis Oliphant (1:38:29.000)

The reason it took a long time,

Lex Fridman (1:38:30.880)

a lot of time, I think it was because one of the things is

Travis Oliphant (1:38:33.320)

there wasn't much to like about Python three.

Lex Fridman (1:38:36.240)

3.0, 3.1, it really wasn't until 3.3.

Travis Oliphant (1:38:40.280)

Like I consider Python 3.3 to be Python 3.0.

Lex Fridman (1:38:43.600)

But it wasn't until Python 3.3

Travis Oliphant (1:38:44.880)

that I felt there's enough stuff in it

Lex Fridman (1:38:47.200)

to make it worth anybody using it, right?

Lex Fridman (1:38:49.800)

And then 3.4 started to be, oh yeah, I want that.

Lex Fridman (1:38:52.600)

And then 3.5 as the matrix multiply operator,

Lex Fridman (1:38:54.880)

and now it's like, okay, we gotta use that.

Lex Fridman (1:38:56.520)

Plus the libraries that started leveraging

Travis Oliphant (1:38:58.400)

some of the features of Python three.

Lex Fridman (1:38:59.600)

Exactly.

Lex Fridman (1:39:00.760)

So it really, the challenge was it was,

Lex Fridman (1:39:03.800)

but it also illustrated a truism that, you know,

Travis Oliphant (1:39:07.400)

when you have inertia,

Lex Fridman (1:39:08.240)

when you have a group of people using something,

Travis Oliphant (1:39:10.480)

it's really hard to move them away from it.

Lex Fridman (1:39:11.960)

You can't just change the world on them.

Lex Fridman (1:39:13.920)

And Python three, you know, made some,

Lex Fridman (1:39:15.440)

I think it fixed some things Guido had always hated.

Travis Oliphant (1:39:17.240)

I don't think he didn't like the fact

Lex Fridman (1:39:18.440)

that print was a statement.

Travis Oliphant (1:39:19.440)

He wanted to make it a function.

Lex Fridman (1:39:20.760)

But in some sense, that's a bit of gratuitous change

Travis Oliphant (1:39:23.200)

to the language.

Lex Fridman (1:39:24.120)

And you could argue, and people have,

Lex Fridman (1:39:27.320)

but one of the challenges was there wasn't enough features

Lex Fridman (1:39:31.520)

and too many just changes without features.

Lex Fridman (1:39:34.960)

And so the empathy for the end user

Lex Fridman (1:39:37.440)

as to why they would switch wasn't there.

Travis Oliphant (1:39:40.480)

I think also it illustrated just the funding realities.

Lex Fridman (1:39:42.960)

Like Python wasn't funded.

Travis Oliphant (1:39:45.040)

Like it was also a project

Lex Fridman (1:39:46.160)

with a bunch of volunteer labor, right?

Travis Oliphant (1:39:48.280)

It had more people, so more volunteer labor,

Lex Fridman (1:39:50.240)

but it was still, it was fun in the sense

Travis Oliphant (1:39:52.240)

that at least Guido had a job.

Lex Fridman (1:39:53.480)

And I've learned some of the behind the scenes on that now

Travis Oliphant (1:39:55.880)

since talking to people who have lived through it

Lex Fridman (1:39:57.840)

and maybe not on air, we can talk about some of that.

Lex Fridman (1:40:00.560)

But it's interesting to see, but Guido had a job,

Lex Fridman (1:40:03.640)

but his full time job wasn't just work on Python.

Travis Oliphant (1:40:07.080)

Like he had other things to do.

Lex Fridman (1:40:08.880)

Just wild.

Lex Fridman (1:40:09.880)

It is wild, isn't it?

Lex Fridman (1:40:10.720)

It's wild how few people are funded.

Travis Oliphant (1:40:13.320)

Yes.

Lex Fridman (1:40:14.160)

And how much impact they have.

Travis Oliphant (1:40:15.200)

Yes.

Lex Fridman (1:40:16.160)

Maybe that's a feature not a bug, I don't know.

Travis Oliphant (1:40:17.920)

Maybe, yes, exactly.

Lex Fridman (1:40:19.080)

At least early on, like it's sort of, I know, yeah.

Travis Oliphant (1:40:21.840)

It's like Olympic athletes are often severely underfunded,

Lex Fridman (1:40:25.160)

but maybe that's what brings out the greatness.

Travis Oliphant (1:40:27.360)

Perhaps, yes, correct.

Lex Fridman (1:40:28.520)

No, exactly.

Travis Oliphant (1:40:29.680)

Maybe this is the essential part of it.

Lex Fridman (1:40:31.880)

Because I do think about that in terms of,

Travis Oliphant (1:40:33.680)

I currently have an incubator for open source startups.

Lex Fridman (1:40:36.200)

Like what I'm trying to do right now

Travis Oliphant (1:40:37.640)

is create the environment I wished had existed

Lex Fridman (1:40:40.480)

when I was leaving academia with NumPy

Lex Fridman (1:40:42.880)

and trying to figure out what to do.

Lex Fridman (1:40:44.120)

I'm trying to create those opportunities and environments.

Travis Oliphant (1:40:46.120)

So, and that's what drives me still,

Lex Fridman (1:40:49.320)

is how do I make the world easier

Lex Fridman (1:40:50.760)

for the open source entrepreneur?

Lex Fridman (1:40:52.600)

So let me stay, I mean, I could probably stay on NumPy

Travis Oliphant (1:40:55.960)

for a long time, but this is fun question.

Lex Fridman (1:41:00.960)

So Andre Kapathy leads the Tesla Autopilot team,

Lex Fridman (1:41:04.680)

and he's also one of the most like legit programmers I know.

Lex Fridman (1:41:10.720)

It's like he builds stuff from scratch a lot,

Lex Fridman (1:41:13.760)

and that's how he builds intuition about how a problem works.

Lex Fridman (1:41:16.200)

He just builds it from scratch, and I always love that.

Lex Fridman (1:41:18.320)

And the primary language he uses is Python

Lex Fridman (1:41:21.320)

for the intuition building.

Lex Fridman (1:41:23.080)

But he posted something on Twitter saying

Lex Fridman (1:41:27.600)

that they got a significant improvement

Travis Oliphant (1:41:31.280)

on some aspect of their like data loading, I think,

Lex Fridman (1:41:35.640)

by switching away from np.square root,

Lex Fridman (1:41:39.840)

so the NumPy's implementation of square root,

Lex Fridman (1:41:42.160)

to math.square root, and then somebody else commented

Travis Oliphant (1:41:44.520)

that you can get even a much greater improvement

Lex Fridman (1:41:48.120)

by using the vanilla Python square root, which is like.

Travis Oliphant (1:41:52.600)

Power 0.5.

Lex Fridman (1:41:53.640)

Power 0.5.

Lex Fridman (1:41:55.200)

And it's fascinating to me, I just wanted to.

Lex Fridman (1:41:58.640)

So that was some shade throwing at some.

Travis Oliphant (1:42:02.080)

No, no, and yes, we're talking about.

Lex Fridman (1:42:04.640)

It's a good way to ask the trade off

Travis Oliphant (1:42:08.080)

between usability and efficiency broadly in NumPy,

Lex Fridman (1:42:12.080)

but also on these specific weird quirks

Travis Oliphant (1:42:14.920)

of like a single function.

Lex Fridman (1:42:16.680)

Yep, so on that point, if you use a NumPy math function

Travis Oliphant (1:42:21.360)

on a scaler, it's gonna be slower

Lex Fridman (1:42:25.000)

than using a Python function on that scaler.

Lex Fridman (1:42:27.960)

But because the math object in NumPy is more complicated,

Lex Fridman (1:42:33.800)

because you can also call that math object on an array.

Lex Fridman (1:42:36.760)

And so effectively, it goes through a similar machine.

Lex Fridman (1:42:39.200)

There aren't enough of the, which you would do

Lex Fridman (1:42:41.840)

and you could do like checks and fast paths.

Lex Fridman (1:42:45.960)

So yeah, if you're basically doing a list,

Travis Oliphant (1:42:48.800)

if you run over a list, in fact,

Lex Fridman (1:42:50.680)

for problems that are less than 1,000,

Travis Oliphant (1:42:53.700)

even maybe 10,000 is probably the,

Lex Fridman (1:42:55.320)

if you're going more than 10,000,

Travis Oliphant (1:42:56.900)

that's where you definitely need to be using arrays.

Lex Fridman (1:42:59.080)

But if you're less than that, and for reading,

Travis Oliphant (1:43:01.200)

if you're doing a reading process

Lex Fridman (1:43:02.760)

and essentially it's not compute bound, it's IO bound.

Lex Fridman (1:43:05.600)

And so you're really taking lists of 1,000 at a time

Lex Fridman (1:43:08.480)

and doing work on it.

Travis Oliphant (1:43:09.540)

Yeah, you could be faster just using Python,

Lex Fridman (1:43:11.680)

straight up Python.

Travis Oliphant (1:43:12.740)

See, but also, and this is the side to the top,

Lex Fridman (1:43:16.640)

there's the fundamental questions

Travis Oliphant (1:43:18.680)

when you look at the long arc of history,

Lex Fridman (1:43:21.240)

it's very possible that np.square root is much faster.

Travis Oliphant (1:43:25.560)

It could be.

Lex Fridman (1:43:26.400)

So like in terms of like, don't worry about it,

Travis Oliphant (1:43:29.480)

it's the evils of over optimization or whatever,

Lex Fridman (1:43:32.420)

all the different quotes around that,

Travis Oliphant (1:43:34.040)

is sometimes obsessing about this particular little quark

Lex Fridman (1:43:39.520)

is not sufficient.

Travis Oliphant (1:43:41.720)

For somebody like, if you're trying to optimize your path,

Lex Fridman (1:43:45.220)

I mean, I agree, premature optimization

Lex Fridman (1:43:47.680)

creates all kinds of challenges, right?

Lex Fridman (1:43:49.320)

Because now, but you may have to do it.

Travis Oliphant (1:43:51.840)

I believe the quote is, it's the root of all evil.

Lex Fridman (1:43:53.880)

It's the root of all evil, right?

Travis Oliphant (1:43:55.560)

Let's give Donald Knuth, I think,

Lex Fridman (1:43:57.040)

or is he more than somebody else?

Travis Oliphant (1:43:59.160)

Well, Doc Knuth is kind of like Mark Twain,

Lex Fridman (1:44:00.800)

people just attribute stuff to him, I don't know.

Lex Fridman (1:44:02.880)

And it's fine because he's brilliant.

Lex Fridman (1:44:04.640)

So, no, I was a LaTeX user myself,

Lex Fridman (1:44:07.640)

and so I have a lot of respect,

Lex Fridman (1:44:09.280)

and he did more than that, of course,

Lex Fridman (1:44:10.820)

but yeah, someone I really appreciate

Lex Fridman (1:44:14.120)

in the computer science space.

Travis Oliphant (1:44:15.640)

Yeah, I don't, I think that's appropriate.

Lex Fridman (1:44:17.080)

There's a lot of little things like that,

Travis Oliphant (1:44:18.320)

where people actually, if you understood it,

Lex Fridman (1:44:20.120)

you go, yeah, of course, that's the case.

Lex Fridman (1:44:22.640)

And the other part, the other part I didn't mention,

Lex Fridman (1:44:25.040)

and Numba was a thing we wrote early on,

Lex Fridman (1:44:27.960)

and I was really excited by Numba

Lex Fridman (1:44:29.040)

because it's something we wanted,

Travis Oliphant (1:44:30.040)

it was a compiler for Python syntax,

Lex Fridman (1:44:32.160)

and I wanted it from the beginning of writing NumPy

Travis Oliphant (1:44:35.440)

because of this function question,

Lex Fridman (1:44:38.280)

like taking, the power of arrays

Travis Oliphant (1:44:41.900)

is really that you can write functions using all of it.

Lex Fridman (1:44:45.120)

It has implicit looping, right?

Lex Fridman (1:44:47.000)

So you don't worry about,

Lex Fridman (1:44:47.840)

I write this n dimensional for loop

Travis Oliphant (1:44:49.200)

with four loops, four, four statements.

Lex Fridman (1:44:51.240)

You just say, oh, big four dimensional array,

Travis Oliphant (1:44:53.600)

I'm gonna do this operation, this plus, this minus,

Lex Fridman (1:44:55.760)

this reduction, and you get this,

Travis Oliphant (1:44:57.680)

it's called vectorization in other areas,

Lex Fridman (1:44:59.560)

but you can basically think at a high level

Lex Fridman (1:45:01.440)

and get massive amounts of computation done

Lex Fridman (1:45:03.640)

with the added benefit of,

Travis Oliphant (1:45:06.200)

oh, it can be paralyzed easily.

Lex Fridman (1:45:08.040)

It can be put in parallel.

Travis Oliphant (1:45:09.040)

You don't have to think about that.

Lex Fridman (1:45:10.000)

In fact, it's worse to go decompose your,

Travis Oliphant (1:45:12.720)

you write the for loops

Lex Fridman (1:45:14.160)

and then try to infer parallelism from for loops.

Travis Oliphant (1:45:16.280)

That's actually a harder problem

Lex Fridman (1:45:17.600)

than to take the array problem

Lex Fridman (1:45:19.640)

and just automatically parallelize that problem.

Lex Fridman (1:45:22.040)

That's what, and so functions in NumPy

Travis Oliphant (1:45:25.320)

are called universal functions, ufuncs.

Lex Fridman (1:45:27.080)

So square root is an example of a ufunk.

Travis Oliphant (1:45:29.000)

There are others, sine, cosine, add, subtract.

Lex Fridman (1:45:32.400)

In fact, one of the first libraries to SciPy

Travis Oliphant (1:45:34.520)

was something called Special

Lex Fridman (1:45:35.520)

where I added Bessel functions

Lex Fridman (1:45:36.920)

and all these special functions that come up in physics

Lex Fridman (1:45:40.240)

and I added them as ufuncs so they could work on arrays.

Lex Fridman (1:45:43.040)

So I understood ufuncs very, very well

Lex Fridman (1:45:44.720)

from day one inside of numeric.

Travis Oliphant (1:45:45.960)

That was one of the things we tried to make better

Lex Fridman (1:45:47.320)

in NumPy was how do they work?

Lex Fridman (1:45:49.120)

Can they do broadcasting?

Lex Fridman (1:45:50.360)

What does broadcasting mean?

Lex Fridman (1:45:51.960)

But one of the problems is, okay,

Lex Fridman (1:45:54.600)

what do I do with a Python scaler?

Lex Fridman (1:45:57.320)

So what happens, the Python scaler gets broadcast

Lex Fridman (1:45:59.800)

to a zero dimensional array

Lex Fridman (1:46:01.320)

and then it goes through the whole same machinery

Lex Fridman (1:46:02.800)

as if it were a 10,000 dimensional array.

Lex Fridman (1:46:05.080)

And then it kind of unpacks the element

Lex Fridman (1:46:07.640)

and then does the addition.

Travis Oliphant (1:46:09.880)

That's not to mention the function it calls

Lex Fridman (1:46:12.600)

in the case of square root

Lex Fridman (1:46:13.640)

is just the clib square root, right?

Lex Fridman (1:46:15.960)

In some cases, like Python's power,

Travis Oliphant (1:46:18.160)

there's some optimizations they're doing

Lex Fridman (1:46:20.360)

that could be faster

Travis Oliphant (1:46:21.520)

than just calling this the clib square root.

Lex Fridman (1:46:23.760)

In the interpreter or in the?

Travis Oliphant (1:46:25.320)

No, in the C code, in the Python runtime.

Lex Fridman (1:46:27.640)

In the Python runtime, so they really optimize it

Lex Fridman (1:46:30.960)

and they have the freedom to do that

Lex Fridman (1:46:32.120)

because they don't have to worry about.

Travis Oliphant (1:46:32.960)

It's just a scaler.

Lex Fridman (1:46:34.080)

It's just a scaler.

Travis Oliphant (1:46:34.920)

Right, they don't have to worry about the fact

Lex Fridman (1:46:36.200)

that, oh, this could be an object with many pieces.

Travis Oliphant (1:46:39.360)

The ufunc machine is also generic

Lex Fridman (1:46:41.080)

in sense that typecasting and broadcasting,

Travis Oliphant (1:46:44.600)

broadcasting's idea of I'm gonna go,

Lex Fridman (1:46:46.160)

I have a zero dimensional array,

Travis Oliphant (1:46:47.360)

I have a scaler with a four dimensional array

Lex Fridman (1:46:49.240)

and I add them.

Travis Oliphant (1:46:50.480)

Oh, I have to kind of coerce the shape of this guy

Lex Fridman (1:46:54.640)

to make it work against the whole four dimensional array.

Lex Fridman (1:46:56.880)

So it's the idea of I can do a one dimensional array

Lex Fridman (1:46:59.680)

against a two dimensional array and have it make sense.

Travis Oliphant (1:47:02.200)

Well, that's what NumPy does is it challenges you

Lex Fridman (1:47:04.040)

to reformulate, rethink your problem

Travis Oliphant (1:47:07.040)

as a multi dimensional array problem

Lex Fridman (1:47:09.080)

versus move away from scalers completely.

Travis Oliphant (1:47:12.640)

Right, exactly, exactly.

Lex Fridman (1:47:14.240)

In fact, that's where some of the edge cases boundaries are

Travis Oliphant (1:47:16.680)

is that, well, they're still there

Lex Fridman (1:47:18.960)

and this is where array scalers are particular.

Lex Fridman (1:47:21.080)

So array scalers are particularly bad

Lex Fridman (1:47:23.120)

in the sense that they were written

Lex Fridman (1:47:24.360)

so that you could optimize the math on them,

Lex Fridman (1:47:26.840)

but that hasn't happened.

Lex Fridman (1:47:29.040)

And so their default is to coerce the array scaler

Lex Fridman (1:47:32.800)

to a zero dimensional array

Lex Fridman (1:47:33.760)

and then use the NumPy machinery.

Lex Fridman (1:47:36.000)

That's what, and you could specialize,

Lex Fridman (1:47:38.200)

but it doesn't happen all the time.

Lex Fridman (1:47:39.960)

So in fact, when we first wrote Numba,

Travis Oliphant (1:47:41.760)

we do comparisons and say, look, it's 1000X speed up.

Lex Fridman (1:47:45.720)

We were lying a little bit in the sense that,

Travis Oliphant (1:47:47.160)

well, first do the 40X slowdown

Lex Fridman (1:47:50.240)

of using the array scalers inside of a loop.

Travis Oliphant (1:47:52.280)

Cause if you used to use Python scalers,

Lex Fridman (1:47:53.560)

you'd already be 10 times faster.

Lex Fridman (1:47:56.200)

But then we would get a hundred times faster

Lex Fridman (1:47:58.080)

over that using just compilation.

Lex Fridman (1:48:00.320)

But what we do is compile the loop

Lex Fridman (1:48:01.600)

from out of the interpreter to machine code.

Lex Fridman (1:48:04.000)

And then that's always been the power of Python

Lex Fridman (1:48:06.280)

is this extensibility so that you can,

Travis Oliphant (1:48:08.280)

cause people say, oh, Python's so slow.

Lex Fridman (1:48:09.680)

Well, sure, if you do all your logic

Travis Oliphant (1:48:11.520)

in the runtime of the Python interpreter, yeah.

Lex Fridman (1:48:13.920)

But the power is that you don't have to.

Travis Oliphant (1:48:15.800)

You write all the logic,

Lex Fridman (1:48:17.260)

what you do in the high level is just high level logic.

Lex Fridman (1:48:19.860)

And the actual calls you're making

Lex Fridman (1:48:21.920)

could be on gigabyte arrays of data.

Lex Fridman (1:48:24.400)

And that's all done at compiled speeds.

Lex Fridman (1:48:26.880)

And the fact that integration is one can happen,

Lex Fridman (1:48:30.320)

but two is separable.

Lex Fridman (1:48:32.420)

That's one of the, the language like Julia says,

Travis Oliphant (1:48:35.240)

we're going to be all in one.

Lex Fridman (1:48:36.380)

You can do all of it together.

Lex Fridman (1:48:37.400)

And then there's, the jury's out, is that possible?

Lex Fridman (1:48:39.880)

I tend to think that you're going to,

Travis Oliphant (1:48:41.760)

there's separate concerns there.

Lex Fridman (1:48:43.280)

You want to precompile.

Travis Oliphant (1:48:44.320)

In fact, generally you will want to precompile your,

Lex Fridman (1:48:47.560)

some of your loops.

Travis Oliphant (1:48:48.400)

Like SciPy is a compilation step.

Lex Fridman (1:48:50.160)

To install SciPy, it takes about two hours.

Travis Oliphant (1:48:53.240)

If you have many machines,

Lex Fridman (1:48:54.080)

maybe you can get it down to one hour.

Lex Fridman (1:48:55.440)

But to compile those libraries takes about, takes a while.

Lex Fridman (1:48:57.920)

You don't want to do that at runtime.

Travis Oliphant (1:48:59.920)

You don't want to do that all the time.

Lex Fridman (1:49:00.800)

You want to have this precompiled binary available

Travis Oliphant (1:49:02.720)

that you're then just linking into.

Lex Fridman (1:49:04.400)

So there's real questions about the whole source code.

Travis Oliphant (1:49:09.040)

Code is, running binary code is more than source code.

Lex Fridman (1:49:11.840)

It's creating object code, it's the linker, it's the loader,

Travis Oliphant (1:49:14.480)

it's the how does that interpret it

Lex Fridman (1:49:15.600)

inside of virtual memory space.

Travis Oliphant (1:49:17.640)

There's a lot of details there that actually

Lex Fridman (1:49:19.160)

I didn't understand for a long time

Travis Oliphant (1:49:20.520)

until I read books on the topic.

Lex Fridman (1:49:23.000)

And it led to, the more you know, the better off you are

Lex Fridman (1:49:27.060)

and you can do more details,

Lex Fridman (1:49:28.440)

but sometimes it helps with abstractions too.

Travis Oliphant (1:49:31.280)

Well, the problem, as we mentioned earlier

Lex Fridman (1:49:33.480)

with abstractions is you kind of sometimes assume

Travis Oliphant (1:49:37.700)

that whoever implemented this thing

Lex Fridman (1:49:41.520)

had your case in mind and found the optimal solution.

Travis Oliphant (1:49:45.000)

Yes.

Lex Fridman (1:49:45.840)

Or like you assume certain things.

Travis Oliphant (1:49:47.320)

I mean, there's a lot of,

Lex Fridman (1:49:48.160)

Correct.

Travis Oliphant (1:49:49.000)

One of the really powerful things to me early on,

Lex Fridman (1:49:52.800)

I mean, it sounds silly to say, but with Python,

Travis Oliphant (1:49:55.480)

probably one of the reasons I fell in love with it

Lex Fridman (1:49:58.440)

is dictionaries.

Travis Oliphant (1:49:59.800)

Yes.

Lex Fridman (1:50:00.920)

So obviously probably most languages

Travis Oliphant (1:50:03.680)

have some mapping concept,

Lex Fridman (1:50:06.440)

but it felt like it was a first class citizen

Lex Fridman (1:50:09.040)

and it was just my brain was able to think in dictionaries.

Lex Fridman (1:50:12.200)

But then there's the thing that I guess I still use

Travis Oliphant (1:50:14.640)

to this day is order dictionaries

Lex Fridman (1:50:16.920)

because that seems like a more natural way

Travis Oliphant (1:50:20.120)

to construct dictionaries.

Lex Fridman (1:50:21.680)

Yeah.

Lex Fridman (1:50:22.520)

And from a computer science perspective,

Lex Fridman (1:50:23.720)

the running time cost is not that significant,

Lex Fridman (1:50:26.000)

but there's a lot of things to understand about dictionaries

Lex Fridman (1:50:30.400)

that the abstraction kind of

Travis Oliphant (1:50:33.800)

doesn't necessarily incentivize you to understand.

Lex Fridman (1:50:37.400)

Right, do you really understand the notion of a hash map

Lex Fridman (1:50:39.400)

and how the dictionary is implemented?

Lex Fridman (1:50:41.080)

But you're right.

Travis Oliphant (1:50:42.080)

Dictionaries are a good example

Lex Fridman (1:50:43.440)

of an abstraction that's powerful.

Lex Fridman (1:50:44.920)

And I agree with you.

Lex Fridman (1:50:46.000)

I agree, I love dictionaries too.

Travis Oliphant (1:50:47.800)

Took me a while to understand that once you do,

Lex Fridman (1:50:49.160)

you realize, oh, they're everywhere.

Lex Fridman (1:50:50.280)

And Python uses them everywhere too.

Lex Fridman (1:50:52.760)

Like it's actually constructed,

Travis Oliphant (1:50:54.240)

one of the foundational things is dictionaries

Lex Fridman (1:50:55.760)

and it does everything with dictionaries.

Lex Fridman (1:50:57.560)

So it is, it's powerful.

Lex Fridman (1:50:58.600)

Order dictionaries came later,

Lex Fridman (1:51:00.160)

but it is very, very powerful.

Lex Fridman (1:51:02.200)

It took me a little while coming

Travis Oliphant (1:51:03.400)

from just the array programming entirely

Lex Fridman (1:51:05.960)

to understand these other objects,

Travis Oliphant (1:51:07.360)

like dictionaries and lists and tuples and binary trees.

Lex Fridman (1:51:11.600)

Like I said, I wasn't a computer scientist,

Travis Oliphant (1:51:13.360)

I studied arrays first.

Lex Fridman (1:51:15.120)

And so I was very array centric.

Lex Fridman (1:51:16.800)

And you realize, oh, these others

Lex Fridman (1:51:17.960)

don't have purposes and value actually.

Travis Oliphant (1:51:21.200)

I agree.

Lex Fridman (1:51:22.040)

There's a friendliness about,

Travis Oliphant (1:51:24.320)

like one way to think about arrays

Lex Fridman (1:51:26.760)

is arrays are just like full of numbers,

Lex Fridman (1:51:31.920)

but to make them accessible to humans

Lex Fridman (1:51:35.000)

and make them less error prone to human users,

Travis Oliphant (1:51:38.700)

sometimes you want to attach names,

Lex Fridman (1:51:41.480)

human interpretable names

Travis Oliphant (1:51:43.120)

that are sticky to those arrays.

Lex Fridman (1:51:44.720)

So that's how you start to think about dictionaries

Travis Oliphant (1:51:47.160)

is you start to convert numbers

Lex Fridman (1:51:50.520)

into something that's human interpretable.

Lex Fridman (1:51:52.120)

And that's actually the tension I've had with NumPy

Lex Fridman (1:51:55.320)

because I've built so much tooling

Travis Oliphant (1:51:58.160)

around human interpretability

Lex Fridman (1:52:02.320)

and also protecting me from a year later

Travis Oliphant (1:52:05.680)

not making the mistakes by being,

Lex Fridman (1:52:07.960)

I wanted to force myself to use English versus numbers.

Travis Oliphant (1:52:12.880)

Yes, so there's a project called Labeled Arrays.

Lex Fridman (1:52:15.680)

Like very early it was recognized that,

Travis Oliphant (1:52:18.040)

oh, we're indexing NumPy with just numbers,

Lex Fridman (1:52:21.320)

all the columns and particularly the dimensions.

Travis Oliphant (1:52:23.640)

I mean, if you have an image,

Lex Fridman (1:52:25.520)

you don't necessarily need to label each column or row,

Lex Fridman (1:52:27.680)

but if you have a lot of images

Lex Fridman (1:52:29.160)

or you have another dimension,

Travis Oliphant (1:52:30.440)

you'd at least like to label the dimension

Lex Fridman (1:52:31.640)

as this is X, this is Y, this is Z,

Travis Oliphant (1:52:33.120)

or this is give us some human meaning

Lex Fridman (1:52:34.640)

or some domain specific meaning.

Travis Oliphant (1:52:36.760)

That was one of the impetuses for Pandas actually

Lex Fridman (1:52:39.680)

was just, oh, we do need to label these things.

Lex Fridman (1:52:43.040)

And Label Array was an attempt to add

Lex Fridman (1:52:45.240)

that like a lighter weight version of that.

Lex Fridman (1:52:47.680)

And there's been, like, that's an example of something

Lex Fridman (1:52:49.360)

I think NumPy could add, could be added to NumPy,

Lex Fridman (1:52:53.080)

but one of the challenges again, how do you fund this?

Lex Fridman (1:52:55.000)

Like I said, one of the tragedies I think is that,

Lex Fridman (1:52:58.280)

so I never had the chance to,

Lex Fridman (1:53:00.240)

I was never paid to work on NumPy, right?

Lex Fridman (1:53:02.360)

So I've always just done it in my spare time,

Lex Fridman (1:53:04.400)

always taken from one thing,

Travis Oliphant (1:53:05.880)

taken from another thing to do it.

Lex Fridman (1:53:07.920)

And at the time, I mean, today,

Travis Oliphant (1:53:09.800)

it would be the wrong day and today,

Lex Fridman (1:53:11.000)

like paying me to work on NumPy now

Travis Oliphant (1:53:12.160)

would not be a good use of effort,

Lex Fridman (1:53:13.480)

but we are finally at Quansight Labs,

Travis Oliphant (1:53:16.640)

I'm actually paying people to work on NumPy and SciPy,

Lex Fridman (1:53:19.440)

which is I'm thrilled with, I'm excited by.

Travis Oliphant (1:53:22.000)

I've wanted to do that.

Lex Fridman (1:53:22.840)

That's what I always wanted to do from day one.

Travis Oliphant (1:53:24.280)

It just took me a while to figure out a mechanism to do that.

Lex Fridman (1:53:27.640)

Even like in the university setting,

Travis Oliphant (1:53:29.680)

respecting that, like pushing students,

Lex Fridman (1:53:33.840)

young minds and young graduate students to contribute

Lex Fridman (1:53:38.000)

and then figuring out financial mechanisms

Lex Fridman (1:53:41.160)

that enable them to contribute

Lex Fridman (1:53:43.280)

and then sort of reward them

Lex Fridman (1:53:45.280)

for their innovative scientific journey,

Travis Oliphant (1:53:48.000)

that would be nice.

Lex Fridman (1:53:49.160)

But then also just a better allocation of resources.

Travis Oliphant (1:53:53.360)

It's 20 year anniversary since 9.11

Lex Fridman (1:53:55.760)

and I was just looking, we spent over $6 trillion

Travis Oliphant (1:53:59.240)

in the Middle East after 9.11 in the various efforts there.

Lex Fridman (1:54:04.560)

And sort of to put politics and all that aside,

Travis Oliphant (1:54:08.040)

it's just, you think about the education system,

Lex Fridman (1:54:10.120)

all the other ways we could have

Travis Oliphant (1:54:11.320)

possibly allocated that money.

Lex Fridman (1:54:14.280)

To me, to take it back,

Travis Oliphant (1:54:16.560)

the amount of impact you would have

Lex Fridman (1:54:21.200)

by allocating a little bit of money to the programmers

Travis Oliphant (1:54:26.360)

that build the tools that run the world is fascinating.

Lex Fridman (1:54:30.600)

It is.

Travis Oliphant (1:54:32.600)

I don't know, I think, again,

Lex Fridman (1:54:34.920)

there is some aspect to being broke

Travis Oliphant (1:54:38.040)

as somewhat of a feature, not a bug,

Lex Fridman (1:54:40.240)

that you make sure that you're valued.

Lex Fridman (1:54:42.320)

But you can still manage that.

Lex Fridman (1:54:43.440)

Right, no, I know.

Lex Fridman (1:54:45.320)

But I don't think that's a big part.

Lex Fridman (1:54:47.040)

So it's like, I think you can have enough money

Lex Fridman (1:54:50.720)

and actually be wealthy while maintaining your values.

Lex Fridman (1:54:53.880)

Agreed, agreed.

Travis Oliphant (1:54:55.520)

There's an old adage that nations that trade together

Lex Fridman (1:54:57.800)

don't go to war together.

Travis Oliphant (1:54:59.440)

I've often thought about nations that code together.

Lex Fridman (1:55:01.680)

Yeah, code together.

Lex Fridman (1:55:02.520)

Right?

Lex Fridman (1:55:03.360)

I love that.

Travis Oliphant (1:55:04.200)

Because one of the things I love about open source

Lex Fridman (1:55:05.360)

is it's global, it's multinational.

Travis Oliphant (1:55:07.880)

Like there aren't national boundaries.

Lex Fridman (1:55:09.160)

One of the challenges with business and open source

Travis Oliphant (1:55:10.760)

is the fact that, well, business is national.

Lex Fridman (1:55:12.800)

Like businesses are entities

Lex Fridman (1:55:13.960)

that are recognized in legal jurisdictions, right?

Lex Fridman (1:55:16.240)

And have laws that are respected in those jurisdictions

Lex Fridman (1:55:18.280)

and hiring, and yet the open source ecosystem

Lex Fridman (1:55:21.320)

is not, it's not there.

Travis Oliphant (1:55:23.040)

Like currently, one of the problems we're solving

Lex Fridman (1:55:25.080)

is hiring people all over the world, right?

Travis Oliphant (1:55:27.200)

Because we, it's a global effort.

Lex Fridman (1:55:29.600)

And I've had the chance to work, and I've loved the chance.

Travis Oliphant (1:55:31.920)

I've never been to like Iran,

Lex Fridman (1:55:35.280)

but I once had a conference

Lex Fridman (1:55:36.800)

where I was able to talk to people there, right?

Lex Fridman (1:55:38.640)

And talk to folks in Pakistan.

Travis Oliphant (1:55:40.920)

I've never been there, but we had a call

Lex Fridman (1:55:44.080)

where there were people there,

Travis Oliphant (1:55:45.320)

like just scientists and normal people.

Lex Fridman (1:55:47.600)

And there's a certain amount of humanizing, right?

Travis Oliphant (1:55:52.640)

That gets away from the,

Lex Fridman (1:55:54.360)

like we often get the memes of society

Travis Oliphant (1:55:56.200)

that bubble up and get discussed,

Lex Fridman (1:55:58.560)

but the memes are not even an accurate reflection

Travis Oliphant (1:56:00.760)

of the reality of what people are.

Lex Fridman (1:56:02.400)

Well, if you look at the major power centers

Travis Oliphant (1:56:05.440)

that are leading to something like cyber war

Lex Fridman (1:56:08.240)

in the next few decades,

Travis Oliphant (1:56:10.000)

it's the United States, it's Russia, and China.

Lex Fridman (1:56:13.320)

And those three countries in particular

Travis Oliphant (1:56:16.080)

have incredible developers.

Lex Fridman (1:56:18.240)

So if they work together, I think that's one way,

Travis Oliphant (1:56:21.360)

the politicians can do their stupid bickering,

Lex Fridman (1:56:23.360)

but like there's a layer of infrastructure, of humanity.

Travis Oliphant (1:56:27.360)

If they collaborate together,

Lex Fridman (1:56:29.400)

that I think can prevent major military conflict,

Travis Oliphant (1:56:34.080)

which would, I think most likely happen at the cyber level

Lex Fridman (1:56:37.840)

versus the actual hot war level.

Travis Oliphant (1:56:39.800)

You're right.

Lex Fridman (1:56:40.640)

You know, I think that's a good prediction.

Travis Oliphant (1:56:43.320)

Nations that code together don't go to war together.

Lex Fridman (1:56:46.560)

Don't go to war together.

Lex Fridman (1:56:47.880)

That's a hope, right?

Lex Fridman (1:56:48.720)

That's one of the philosophical hopes, but yeah.

Lex Fridman (1:56:52.360)

So you mentioned the project of Numba,

Lex Fridman (1:56:55.640)

which is fascinating.

Lex Fridman (1:56:58.520)

So from the early days,

Lex Fridman (1:56:59.720)

there was kind of a pushback on Python that it's not fast.

Travis Oliphant (1:57:04.560)

You know, you see C plus,

Lex Fridman (1:57:05.520)

if you wanna write something that's fast,

Travis Oliphant (1:57:06.920)

you use C plus plus.

Lex Fridman (1:57:08.240)

If you wanna write something that's usable and friendly,

Lex Fridman (1:57:11.320)

but slow, you use Python.

Lex Fridman (1:57:13.240)

And so what is Numba?

Lex Fridman (1:57:15.840)

What is its goal?

Lex Fridman (1:57:16.800)

How does it work?

Travis Oliphant (1:57:17.640)

Great, yeah.

Lex Fridman (1:57:18.480)

Yes, that's what the argument.

Lex Fridman (1:57:19.760)

And the reality was people would write high level coding

Lex Fridman (1:57:22.440)

and use compiled code,

Lex Fridman (1:57:23.440)

but there's still user stories, use cases,

Lex Fridman (1:57:25.240)

where you want to write Python,

Lex Fridman (1:57:27.440)

but then have it still be fast.

Lex Fridman (1:57:28.880)

You still need to write a for loop.

Travis Oliphant (1:57:30.720)

Like before Numba, it was always don't write a for loop.

Lex Fridman (1:57:33.920)

You know, write it in a vectorized way,

Travis Oliphant (1:57:35.800)

you know, put it in an array.

Lex Fridman (1:57:37.240)

And often that can make a memory trade off.

Travis Oliphant (1:57:39.640)

Like quite often you can do it,

Lex Fridman (1:57:41.080)

but then you make maybe use more memory

Travis Oliphant (1:57:42.720)

because you have to build this array of data

Lex Fridman (1:57:44.920)

that you don't necessarily need all the time.

Lex Fridman (1:57:46.680)

So Numba was, it started from a desire to have

Lex Fridman (1:57:50.960)

kind of a vectorized that worked.

Travis Oliphant (1:57:52.840)

A vectorized was a tool in NumPy, it was released.

Lex Fridman (1:57:56.260)

You give it a Python function

Lex Fridman (1:57:57.800)

and it gave you a universal function,

Lex Fridman (1:57:59.680)

a ufunc that would work on arrays.

Lex Fridman (1:58:01.120)

So you get the function that just worked on a scaler.

Lex Fridman (1:58:03.640)

Like you could make a,

Travis Oliphant (1:58:04.880)

like the classic case was a simple function

Lex Fridman (1:58:07.280)

that an if then statement in it.

Lex Fridman (1:58:08.280)

So sine X over X function, sync function.

Lex Fridman (1:58:12.160)

If X equals zero, return one, otherwise do sine X over X.

Travis Oliphant (1:58:16.080)

The challenge is you don't want that loop

Lex Fridman (1:58:17.760)

peg one in Python.

Lex Fridman (1:58:18.720)

So you want a compiled version of that,

Lex Fridman (1:58:21.480)

but the ufunc, the vectorized in NumPy

Travis Oliphant (1:58:23.160)

would just give you a Python function.

Lex Fridman (1:58:24.840)

So it would take the array of numbers

Lex Fridman (1:58:26.720)

and at every call do a loop back into Python.

Lex Fridman (1:58:29.560)

So it was very slow.

Travis Oliphant (1:58:30.440)

It gave you the appearance of a ufunc,

Lex Fridman (1:58:31.800)

but it was very slow.

Lex Fridman (1:58:32.840)

So I always wanted a vectorized

Lex Fridman (1:58:34.600)

that would take that Python scaler function

Lex Fridman (1:58:36.280)

and produce a ufunc working on binary native code.

Lex Fridman (1:58:39.480)

So in fact, I had somebody work on that with PyPy

Lex Fridman (1:58:42.800)

and see if PyPy could be used to produce a ufunc like that

Lex Fridman (1:58:45.640)

early on in 2009 or something like that, 2010.

Travis Oliphant (1:58:50.560)

They didn't work that well.

Lex Fridman (1:58:51.480)

It was kind of pretty bulky.

Lex Fridman (1:58:52.880)

But in 2012, Peter and I had just started Anaconda.

Lex Fridman (1:58:57.000)

We had, I just, I'd learned to raise money.

Travis Oliphant (1:59:00.680)

That's a different topic,

Lex Fridman (1:59:01.640)

but I'd learned to raise money from friends, family,

Lex Fridman (1:59:04.640)

and fools, as they say.

Lex Fridman (1:59:05.960)

And.

Travis Oliphant (1:59:06.800)

That's a good line.

Lex Fridman (1:59:09.840)

Oh, that's a good line.

Travis Oliphant (1:59:11.200)

But, so we were trying to do something.

Lex Fridman (1:59:13.440)

We were trying to change the world.

Travis Oliphant (1:59:14.680)

Peter and I are super ambitious.

Lex Fridman (1:59:15.840)

We wanted to make array computing

Lex Fridman (1:59:17.600)

and we had ideas for really what's still,

Lex Fridman (1:59:19.480)

it's still the energy right now.

Lex Fridman (1:59:20.640)

How do you do at scale data science?

Lex Fridman (1:59:23.520)

And we had a bunch of ideas there, but one of them,

Travis Oliphant (1:59:25.840)

I had just talked to people about LLVM

Lex Fridman (1:59:27.720)

and I was like, there's a way to do this.

Travis Oliphant (1:59:30.040)

I just, I went, I heard about my friend Dave Beasley

Lex Fridman (1:59:32.600)

at a compiler course.

Lex Fridman (1:59:33.920)

So I was looking at compilers like,

Lex Fridman (1:59:35.560)

and I realized, oh, this is what you do.

Lex Fridman (1:59:37.640)

And so I wrote a version of Numba

Lex Fridman (1:59:40.040)

that just basically mapped Python bytecode to LLVM.

Travis Oliphant (1:59:45.640)

Nice.

Lex Fridman (1:59:46.480)

Right, so, and the first version is like, this works

Lex Fridman (1:59:49.200)

and it produces code that's fast.

Lex Fridman (1:59:50.840)

This is cool for, you know,

Travis Oliphant (1:59:51.960)

obviously a reduced subset of Python.

Lex Fridman (1:59:53.440)

I didn't support all the Python language.

Travis Oliphant (1:59:55.360)

There had been efforts to speed up Python in the past,

Lex Fridman (1:59:57.480)

but those efforts were, I would say,

Travis Oliphant (1:59:59.200)

not from the array computing perspective,

Lex Fridman (20:02.540)

that I would say APLJ was another version that was,

Lex Fridman (20:06.580)

what it did is not have the glyphs,

Lex Fridman (20:08.340)

just have short characters,

Lex Fridman (20:09.700)

but still a Latin keyboard could type them.

Lex Fridman (20:11.740)

And then numeric inherited from that

Travis Oliphant (20:14.540)

in terms of let's add arrays plus broadcasting

Lex Fridman (20:17.660)

plus methods, reduction,

Travis Oliphant (20:19.700)

even some of the language like rank is a concept

Lex Fridman (20:21.780)

that was in Python and is still in Python

Lex Fridman (20:24.660)

for the number of dimensions, right?

Lex Fridman (20:27.180)

That's different than say the rank of a matrix

Travis Oliphant (20:29.460)

which people think of as well.

Lex Fridman (20:31.140)

So it came from that tradition,

Lex Fridman (20:33.060)

but NumPy is a very pragmatic, practical tool.

Lex Fridman (20:37.980)

NumPy inherited from numeric

Lex Fridman (20:39.260)

and we can get to where NumPy came from

Lex Fridman (20:40.820)

which is the current array,

Travis Oliphant (20:43.340)

at least current as of 2015, 2017.

Lex Fridman (20:46.100)

Now there's a ton of them over the past two or three years.

Travis Oliphant (20:49.320)

We can get into that too.

Lex Fridman (20:50.320)

So if we just linger on the early days

Lex Fridman (20:52.780)

of what was your favorite feature of Python?

Lex Fridman (20:56.220)

Do you remember like what?

Lex Fridman (20:58.020)

So it's so interesting to linger on like the,

Lex Fridman (21:02.260)

what really makes you connect with a language?

Travis Oliphant (21:06.300)

I'm not sure it's obvious to introspect that.

Lex Fridman (21:09.400)

No, it isn't.

Lex Fridman (21:10.240)

And I've thought about that at some length.

Lex Fridman (21:12.860)

I think definitely the fact that I could read it later,

Travis Oliphant (21:16.460)

that I could use it productively

Lex Fridman (21:18.140)

without becoming an expert.

Travis Oliphant (21:19.820)

Other language I had to put more effort into.

Lex Fridman (21:22.180)

That's like an empirical observation.

Travis Oliphant (21:23.940)

Like you're not analyzing any one aspect of the language.

Lex Fridman (21:26.500)

It just seems time after time when you look back,

Travis Oliphant (21:29.460)

it's somehow readable.

Lex Fridman (21:30.580)

It's somehow readable.

Travis Oliphant (21:31.420)

Then it was sort of, I could take executable English

Lex Fridman (21:35.380)

and translate it to Python more easily.

Travis Oliphant (21:36.820)

Like I didn't have to go, there was no translation layer.

Lex Fridman (21:39.760)

As an engineer or as a scientist,

Travis Oliphant (21:41.580)

I could think about what I wanted to do.

Lex Fridman (21:43.240)

And then the syntax wasn't that far behind it, right?

Travis Oliphant (21:46.780)

Now there are some warts there still.

Lex Fridman (21:49.220)

It wasn't perfect.

Travis Oliphant (21:50.600)

Like there's some areas where I'm like,

Lex Fridman (21:51.440)

ah, it'd be better if this were different

Travis Oliphant (21:52.820)

or if this were different.

Lex Fridman (21:54.380)

Some of those things got added to the language too.

Travis Oliphant (21:56.580)

I was really grateful for some of the early pioneers

Lex Fridman (21:58.580)

in the Python ecosystem back,

Travis Oliphant (22:00.220)

because Python got written in 91.

Lex Fridman (22:01.900)

That's when the first version came out.

Lex Fridman (22:03.140)

But Guido was very open to users.

Lex Fridman (22:06.540)

And one of the sets of users were people like Jim Huganen

Lex Fridman (22:08.660)

and David Asher and Paul Dubois and Conrad Hinson.

Lex Fridman (22:13.460)

These were people that were on the main list.

Lex Fridman (22:15.380)

And they were just asking for things like,

Lex Fridman (22:16.860)

hey, we really should have complex numbers in this language.

Lex Fridman (22:19.220)

So let's, you know, there's a J, there's a one J, right?

Lex Fridman (22:22.540)

And the fact that they went the engineering route of J

Travis Oliphant (22:24.340)

is interesting.

Lex Fridman (22:26.660)

I don't think that's entirely favoring engineers.

Travis Oliphant (22:28.620)

I think it's because I is so often used

Lex Fridman (22:30.460)

as the index of a for loop.

Lex Fridman (22:32.100)

So I think that's actually why.

Lex Fridman (22:34.260)

Probably, I mean, there's a pragmatic aspect.

Lex Fridman (22:36.740)

But the fact that complex numbers were there, I love that.

Lex Fridman (22:39.100)

The fact that I could write in the array constructs

Lex Fridman (22:41.460)

and that reduction was there,

Lex Fridman (22:42.820)

very simple to write summations and broadcasting was there.

Travis Oliphant (22:46.540)

I could do addition of whole arrays.

Lex Fridman (22:49.440)

So that was cool.

Travis Oliphant (22:50.380)

Those are some things I loved about it.

Lex Fridman (22:52.660)

I don't know what to start talking to you about

Travis Oliphant (22:54.820)

because you've created so many incredible projects

Lex Fridman (22:57.860)

that basically changed the whole landscape of programming.

Lex Fridman (23:00.180)

But okay, let's start with,

Lex Fridman (23:02.380)

let's go chronologically with SciPy.

Lex Fridman (23:06.060)

You created SciPy over two decades ago now?

Lex Fridman (23:09.100)

Yes, yes, I love to talk about SciPy.

Travis Oliphant (23:11.140)

SciPy was really my baby.

Lex Fridman (23:12.980)

What is it?

Lex Fridman (23:14.420)

What was its goal?

Lex Fridman (23:15.420)

What is its goal?

Lex Fridman (23:16.420)

How does it work?

Lex Fridman (23:17.260)

Yeah, fantastic.

Lex Fridman (23:18.100)

So SciPy was effectively, here I am using Python

Lex Fridman (23:21.580)

to do stuff that I previously used MATLAB to use.

Lex Fridman (23:24.980)

And I was using numeric, which is an array library

Lex Fridman (23:26.860)

that made a lot of it possible.

Lex Fridman (23:28.300)

But there's things that were missing.

Lex Fridman (23:29.900)

Like I didn't have an ordinary differential equation solver

Lex Fridman (23:32.100)

I could just call, right?

Lex Fridman (23:33.460)

I didn't have integration.

Travis Oliphant (23:35.260)

Hey, I wanted to integrate this function.

Lex Fridman (23:37.180)

Okay, well, I don't have just a function

Travis Oliphant (23:38.780)

I can call to do that.

Lex Fridman (23:40.580)

These are things I remember being critical things

Travis Oliphant (23:42.540)

that I was missing.

Lex Fridman (23:43.700)

Optimization.

Travis Oliphant (23:44.580)

I just wanna pass a function to an optimizer

Lex Fridman (23:46.780)

and have it tell me what the optimal value is.

Travis Oliphant (23:50.100)

Those are things I'm like, well,

Lex Fridman (23:51.100)

why don't we just write a library that adds these tools?

Lex Fridman (23:54.340)

And I started to post on the mailing list

Lex Fridman (23:55.740)

and there'd previously been, people have discussed,

Travis Oliphant (23:58.100)

I remember Conrad Henson saying,

Lex Fridman (23:59.140)

wouldn't it be great if we had this optimizer library

Travis Oliphant (24:00.980)

or David Ashwood say this stuff.

Lex Fridman (24:02.580)

And I'm a ambitious, ambitious is the wrong word,

Travis Oliphant (24:06.940)

an eager and probably more time than sense.

Lex Fridman (24:11.340)

I was a poor graduate student.

Travis Oliphant (24:13.620)

My wife thinks I'm working on my PhD and I am,

Lex Fridman (24:15.860)

but part of the PhD that I loved

Travis Oliphant (24:17.220)

was the fact that it's exploratory.

Lex Fridman (24:19.180)

You're not just taking orders,

Travis Oliphant (24:21.540)

fulfilling a list of things to do,

Lex Fridman (24:23.500)

you're trying to figure out what to do.

Lex Fridman (24:25.740)

And so I thought, well, I'm running tools

Lex Fridman (24:27.900)

for my own use and a PhD,

Lex Fridman (24:29.140)

so I'll just start this project.

Lex Fridman (24:32.140)

And so in 99, 98 was when I first started

Travis Oliphant (24:34.940)

to write libraries for Python.

Lex Fridman (24:36.620)

Definitely when I fell in love with Python 98,

Travis Oliphant (24:38.260)

I thought, oh, well, there's just a few things missing.

Lex Fridman (24:39.740)

Like, oh, I need a reader to read DICOM files.

Travis Oliphant (24:42.700)

I was in medical imaging and DICOM was a format

Lex Fridman (24:44.580)

that I want to be able to load that into Python.

Lex Fridman (24:46.940)

Okay, how do I write a reader for that?

Lex Fridman (24:48.180)

So I wrote something called, it was an IO package, right?

Lex Fridman (24:51.700)

And that was my very first extension module, which is C.

Lex Fridman (24:55.140)

So I wrote C code to extend Python

Lex Fridman (24:57.060)

so that in Python I could write things more easily.

Lex Fridman (24:59.660)

That combination kind of hooked me.

Travis Oliphant (25:02.260)

It was the idea that I could,

Lex Fridman (25:03.300)

here's this powerful tool I can use as a scripting language

Lex Fridman (25:05.700)

and a high level language to think about,

Lex Fridman (25:07.460)

but that I can extend easily, easily in C,

Travis Oliphant (25:11.420)

easily for me because I knew enough C.

Lex Fridman (25:13.780)

And then Guido had written a link.

Travis Oliphant (25:15.260)

I mean, the only, the hard part of extending Python

Lex Fridman (25:17.220)

was something called the way memory management networks,

Lex Fridman (25:19.500)

and you have to do reference counting.

Lex Fridman (25:21.060)

And so there's a tracking of reference counting

Travis Oliphant (25:23.820)

you have to do manually.

Lex Fridman (25:25.500)

And if you don't, you have memory leaks.

Lex Fridman (25:27.500)

And so that's hard.

Lex Fridman (25:29.020)

Plus then C, you know, it's just much more,

Travis Oliphant (25:31.020)

you have to put more effort into it.

Lex Fridman (25:32.180)

It's not just, I have to now think about pointers

Lex Fridman (25:34.700)

and I have to think about stuff that is different.

Lex Fridman (25:37.620)

I have to kind of,

Travis Oliphant (25:38.460)

you're like putting a new cartridge in your brain.

Lex Fridman (25:40.620)

Like, okay, I'm thinking about MRI.

Travis Oliphant (25:42.380)

Now I'm thinking about programming.

Lex Fridman (25:43.580)

And there are distinct modules

Travis Oliphant (25:45.340)

you end up having to think about.

Lex Fridman (25:46.620)

So it's harder.

Lex Fridman (25:47.460)

And when I was just in Python,

Lex Fridman (25:48.300)

I could just think about MRI and high level writing,

Lex Fridman (25:51.500)

but I could do that.

Lex Fridman (25:52.340)

And that kind of, I liked it.

Travis Oliphant (25:54.020)

I found that to be enjoyable and fun.

Lex Fridman (25:55.780)

And so I ended up, oh,

Travis Oliphant (25:57.220)

well, let me just add a bunch of stuff to Python

Lex Fridman (25:59.020)

to do integration.

Travis Oliphant (26:00.580)

Well, and the cool thing is,

Lex Fridman (26:01.660)

is that the power of the internet,

Travis Oliphant (26:03.060)

just looking around and I found,

Lex Fridman (26:04.300)

oh, there's this NetLive,

Travis Oliphant (26:06.300)

which has hundreds of 4chan routines

Lex Fridman (26:08.860)

that people have written in the 60s and the 70s and the 80s

Travis Oliphant (26:12.260)

in 4chan 77, fortunately, it wasn't 4chan 16.

Lex Fridman (26:14.900)

So it had been ported to 4chan 77.

Lex Fridman (26:18.100)

And 4chan 77 is actually a really great language.

Lex Fridman (26:21.660)

4chan 90 probably is my favorite 4chan

Travis Oliphant (26:24.100)

because it's also, it's got complex numbers,

Lex Fridman (26:26.100)

got arrays and it's pretty high level.

Travis Oliphant (26:27.700)

Now, the problem with it

Lex Fridman (26:28.980)

is you'd never want to write a program in 4chan 90

Travis Oliphant (26:31.020)

or 4chan 77,

Lex Fridman (26:32.260)

but it's totally fine to write a subroutine in, right?

Lex Fridman (26:34.900)

And so, and then 4chan kind of got a little off course

Lex Fridman (26:37.660)

when they tried to compete with C++.

Lex Fridman (26:39.060)

But at the time,

Lex Fridman (26:40.580)

I just want libraries to do something like,

Travis Oliphant (26:42.340)

oh, here's an ordinary differential equation.

Lex Fridman (26:43.940)

Here's integration.

Travis Oliphant (26:44.900)

Here's runge cut integration.

Lex Fridman (26:46.780)

Already done.

Travis Oliphant (26:47.620)

I don't have to think about that algorithm.

Lex Fridman (26:48.780)

I mean, you could,

Lex Fridman (26:49.620)

but it's nice to have somebody who's already done one

Lex Fridman (26:51.020)

and tested it.

Lex Fridman (26:51.860)

And so I sort of started this journey in 98, really.

Lex Fridman (26:55.060)

If you look back at the mailing list,

Travis Oliphant (26:55.980)

there's sort of this productive era of me

Lex Fridman (26:59.660)

writing an extension module

Travis Oliphant (27:01.100)

to connect runge cut integration to Python

Lex Fridman (27:04.580)

and making an ordinary differential equation solver.

Lex Fridman (27:06.660)

And then releasing that as a package.

Lex Fridman (27:09.140)

So we could call ODE pack, I think I called it then.

Travis Oliphant (27:11.820)

Quad pack.

Lex Fridman (27:12.660)

And then I just made these packages.

Travis Oliphant (27:14.420)

Eventually that became multipack

Lex Fridman (27:16.260)

because they're originally modular.

Travis Oliphant (27:17.580)

You can install them separately.

Lex Fridman (27:19.140)

But a massive problem in Python

Travis Oliphant (27:20.700)

was actually just getting your stuff installed.

Lex Fridman (27:23.420)

At the time, releasing software for me,

Lex Fridman (27:25.820)

like today it's people think, what does that mean?

Lex Fridman (27:27.580)

Well, then it meant some poorly written webpage.

Travis Oliphant (27:30.780)

I had some bad webpage up and I put a tarball,

Lex Fridman (27:33.100)

just a GZIP tarball of source code.

Travis Oliphant (27:35.780)

That was the release.

Lex Fridman (27:37.140)

But okay, can we just stand that?

Travis Oliphant (27:39.180)

Because the community aspect

Lex Fridman (27:43.060)

of creating the package and sharing that, that's rare.

Travis Oliphant (27:47.820)

That, to have, to both have the, at that time,

Lex Fridman (27:50.940)

so like the raw.

Travis Oliphant (27:51.780)

Yeah, it was pretty early, yeah.

Lex Fridman (27:52.740)

Oh, well, not rare.

Travis Oliphant (27:54.660)

Maybe you can correct me on this,

Lex Fridman (27:57.020)

but it seems like in the scientific community,

Lex Fridman (27:59.660)

so many people, you were basically solving the problems

Lex Fridman (28:02.420)

you needed to solve to process the particular application,

Travis Oliphant (28:07.100)

the data that you need.

Lex Fridman (28:08.540)

And to also have the mind

Travis Oliphant (28:10.900)

that I'm going to make this usable for others, that's.

Lex Fridman (28:15.340)

I would say I was inspired.

Travis Oliphant (28:16.500)

I'd been inspired by Linux,

Lex Fridman (28:18.060)

been inspired by Linus and him making his code available.

Lex Fridman (28:21.820)

And I was starting to use Linux at the time.

Lex Fridman (28:23.260)

And I went, this is cool.

Lex Fridman (28:24.460)

So I'd kind of been previously primed that way.

Lex Fridman (28:27.060)

And generally I was into science

Travis Oliphant (28:29.180)

because I liked the sharing notion.

Lex Fridman (28:30.980)

I liked the idea of, hey, let's,

Travis Oliphant (28:32.660)

if collectively we build knowledge and share it,

Lex Fridman (28:34.780)

we can all be better off.

Travis Oliphant (28:35.740)

Okay, so you want to energize by that idea.

Lex Fridman (28:37.420)

So I was energized by that idea already, right?

Lex Fridman (28:39.540)

And I can't deny that I was.

Lex Fridman (28:40.940)

I'm sort of had this very,

Travis Oliphant (28:42.900)

I liked that part of science, that part of sharing.

Lex Fridman (28:45.700)

And then all of a sudden, oh, wait, here's something.

Lex Fridman (28:47.300)

And here's something I could do.

Lex Fridman (28:49.940)

And then I slowly over years learned how to share better

Lex Fridman (28:52.780)

so that you could actually engage more people faster.

Lex Fridman (28:55.100)

One of the key things was actually giving people a binary

Lex Fridman (28:57.100)

they could install, right?

Lex Fridman (28:58.980)

So that it wasn't just your source code, good luck.

Travis Oliphant (29:01.460)

Compile this and then.

Lex Fridman (29:02.660)

It's compiled, ready to install, just, you know.

Lex Fridman (29:05.180)

So in fact, a lot of the journey from 98,

Lex Fridman (29:07.380)

even through 2012 when I started Anaconda was about that.

Travis Oliphant (29:10.780)

Like it's why, you know, it's really the key

Lex Fridman (29:13.260)

as to why a scientist with dreams of doing MRI research

Travis Oliphant (29:17.460)

ended up starting a software company

Lex Fridman (29:19.500)

that installs software.

Travis Oliphant (29:22.260)

I work with a few folks now that don't program

Lex Fridman (29:26.700)

like on the creative side and the video side,

Travis Oliphant (29:28.580)

the audio side.

Lex Fridman (29:29.620)

And because my whole life is running on scripts,

Travis Oliphant (29:32.500)

I have to try to get them,

Lex Fridman (29:34.020)

I'm having all the task of teaching them

Lex Fridman (29:35.900)

how to do Python enough to run the scripts.

Lex Fridman (29:39.220)

And so I've been actually facing this,

Travis Oliphant (29:40.820)

whether it's Anaconda or some with the task of

Lex Fridman (29:44.220)

how do I minimally explain basically to my mom

Lex Fridman (29:46.780)

how to write a Python script.

Lex Fridman (29:48.900)

And it's an interesting challenge.

Travis Oliphant (29:50.500)

I have to, it's a to do item for me to figure out like,

Lex Fridman (29:53.020)

what is the minimal amount of information I have to teach?

Lex Fridman (29:56.340)

What are the tools you use that one, you enjoy it,

Lex Fridman (29:59.700)

two, you're effective at it.

Travis Oliphant (2:00:00.840)

not from the perspective of wanting to produce

Lex Fridman (2:00:02.160)

a vectorized improvement.

Travis Oliphant (2:00:03.560)

They were from the perspective of speeding up

Lex Fridman (2:00:05.120)

the runtime of Python, which is fundamentally hard

Travis Oliphant (2:00:07.520)

because Python allows for some constructs

Lex Fridman (2:00:10.520)

that aren't, you can't speed up.

Travis Oliphant (2:00:12.160)

Like it's this generic, you know, when it does this variable.

Lex Fridman (2:00:15.560)

So I, from the start, did not try to replicate

Travis Oliphant (2:00:17.720)

Python's semantics entirely.

Lex Fridman (2:00:20.280)

I said, I'm gonna take a subset of the Python syntax

Lex Fridman (2:00:23.000)

and let people write syntax in Python,

Lex Fridman (2:00:25.080)

but it's kind of a new language really.

Lex Fridman (2:00:27.440)

So it's almost like four loops, like focusing on four loops.

Lex Fridman (2:00:30.480)

Four loops, scalar arithmetic, you know, typed,

Travis Oliphant (2:00:34.400)

you know, really typed language, a typed subset.

Lex Fridman (2:00:38.280)

That was the key.

Travis Oliphant (2:00:39.360)

So, but we wanted to add inference of types.

Lex Fridman (2:00:41.880)

So you didn't have to spell all the types out

Travis Oliphant (2:00:43.400)

because when you call a function,

Lex Fridman (2:00:45.840)

so Python is typed, it's just dynamically typed.

Lex Fridman (2:00:48.040)

So you don't tell it what the types are,

Lex Fridman (2:00:49.360)

but when it runs, every time an object runs,

Travis Oliphant (2:00:52.080)

there's a type for the variables.

Lex Fridman (2:00:53.360)

You know what it is.

Lex Fridman (2:00:54.560)

And so that was the design goals of Numba

Lex Fridman (2:00:56.800)

were to make it possible to write functions

Travis Oliphant (2:00:59.200)

that could be compiled and have them used for NumPy arrays.

Lex Fridman (2:01:03.440)

Like they needed to support NumPy arrays.

Lex Fridman (2:01:05.520)

And so how does it work?

Lex Fridman (2:01:07.040)

Do you add a comment within Python that tells it to do,

Lex Fridman (2:01:10.200)

like how do you help out the compiler?

Lex Fridman (2:01:11.880)

Yeah, so there isn't much actually.

Travis Oliphant (2:01:15.860)

You don't, it's kind of magical in the sense

Lex Fridman (2:01:17.740)

that it just looks at the type of the objects

Lex Fridman (2:01:19.600)

and then it's typed inference to determine

Lex Fridman (2:01:21.320)

any other variables it needs.

Lex Fridman (2:01:23.320)

And then it was also, because we had a use case

Lex Fridman (2:01:26.080)

that could work early.

Travis Oliphant (2:01:28.280)

Like one of the challenges of any kind of new development

Lex Fridman (2:01:30.700)

is if you have something that to make it work,

Travis Oliphant (2:01:32.280)

it was gonna take you a long time,

Lex Fridman (2:01:34.200)

it's really hard to get out off the ground.

Travis Oliphant (2:01:35.960)

If you have a project where there's some incremental story,

Lex Fridman (2:01:39.200)

it can start working today and solve a problem,

Travis Oliphant (2:01:42.300)

then you can start getting it out there, getting feedback.

Lex Fridman (2:01:44.640)

Because Numba today, now Numba is nine years old today,

Lex Fridman (2:01:48.160)

the first two, three versions were not great, right?

Lex Fridman (2:01:52.120)

But they solved a problem and some people could try it

Lex Fridman (2:01:54.120)

and we could get some feedback on it.

Lex Fridman (2:01:55.560)

Not great in that it was very focused.

Travis Oliphant (2:01:57.520)

Very fragile, the subset it would actually compile

Lex Fridman (2:02:02.000)

was small and so if you wrote Python code

Lex Fridman (2:02:04.320)

and said, so the way it worked is you write a function

Lex Fridman (2:02:06.880)

and you say at JIT, use decorators.

Lex Fridman (2:02:09.000)

So decorators, just these little constructs

Lex Fridman (2:02:11.040)

let you decorate code with an at and then a name.

Travis Oliphant (2:02:15.040)

The at JIT would take your Python function

Lex Fridman (2:02:17.760)

and actually just compile it and replace the Python function

Travis Oliphant (2:02:20.240)

with another function that interacts

Lex Fridman (2:02:23.200)

with this compiled function.

Lex Fridman (2:02:24.920)

And it would just do that and we went from Python bytecode

Lex Fridman (2:02:28.480)

then we went to AST.

Travis Oliphant (2:02:29.400)

I mean, writing compilers actually,

Lex Fridman (2:02:31.200)

I learned a lot about why computer science

Travis Oliphant (2:02:32.940)

is taught the way it is because compilers

Lex Fridman (2:02:35.560)

can be hard to write.

Travis Oliphant (2:02:36.640)

They use tree structures, they use all the concepts

Lex Fridman (2:02:39.080)

of computer science that are needed.

Travis Oliphant (2:02:40.520)

It's actually hard to, it's easy to write a compiler

Lex Fridman (2:02:44.600)

and then have it be spaghetti code.

Travis Oliphant (2:02:46.000)

Like the passes become challenging

Lex Fridman (2:02:47.600)

and we ended up with three versions of Numba, right?

Travis Oliphant (2:02:49.940)

Numba got written three times.

Lex Fridman (2:02:51.540)

What programming language is Numba written in?

Travis Oliphant (2:02:55.560)

Python.

Lex Fridman (2:02:56.440)

Wait, okay.

Travis Oliphant (2:02:57.440)

Yeah, Python.

Lex Fridman (2:02:58.640)

So.

Lex Fridman (2:03:00.040)

Really?

Lex Fridman (2:03:00.860)

That's fascinating.

Travis Oliphant (2:03:01.700)

Yeah, so Python, but then the whole goal of Numba

Lex Fridman (2:03:03.520)

is to translate Python bytecode to LLVM.

Lex Fridman (2:03:07.480)

And so LLVM actually does the code generation.

Lex Fridman (2:03:09.400)

In fact, a lot of times they'd say,

Travis Oliphant (2:03:10.780)

yeah, it's super easy to write a compiler

Lex Fridman (2:03:12.780)

if you're not writing the parser nor the code generator.

Lex Fridman (2:03:15.880)

Right?

Lex Fridman (2:03:16.720)

So for people who don't know, LLVM is a compiler itself.

Lex Fridman (2:03:19.440)

So your compiler.

Lex Fridman (2:03:20.360)

Yeah, it's really badly named low level virtual machine,

Travis Oliphant (2:03:22.680)

which that part of it is not used.

Lex Fridman (2:03:24.480)

It's really low level.

Travis Oliphant (2:03:25.320)

Chris, he doesn't mean that.

Lex Fridman (2:03:26.160)

Yeah, love Chris.

Lex Fridman (2:03:29.280)

But the name makes you imply that the virtual machine

Lex Fridman (2:03:31.640)

is what it's all about.

Travis Oliphant (2:03:32.480)

It's actually the IR and the library,

Lex Fridman (2:03:34.520)

the code generation.

Travis Oliphant (2:03:36.000)

That's the real beauty of it.

Lex Fridman (2:03:37.680)

The fact that, what I love about LLVM

Travis Oliphant (2:03:39.360)

was the fact that it was a plateau you could collaborate on.

Lex Fridman (2:03:43.200)

Right?

Travis Oliphant (2:03:44.040)

Instead of the internals of GCC

Lex Fridman (2:03:45.880)

or the internals of the Intel compiler,

Lex Fridman (2:03:47.440)

or like how do I extend that?

Lex Fridman (2:03:49.120)

And it was a place we could collaborate.

Lex Fridman (2:03:51.020)

And we were early.

Lex Fridman (2:03:52.400)

I mean, people had started before.

Travis Oliphant (2:03:54.000)

It's a slow compiler.

Lex Fridman (2:03:55.240)

Like it's not a fast compiler.

Lex Fridman (2:03:56.840)

So for some kind of JITs,

Lex Fridman (2:03:59.520)

like JITs are common in language

Travis Oliphant (2:04:01.040)

because one, every browser has a JavaScript JIT.

Lex Fridman (2:04:04.760)

It does real time compilation

Travis Oliphant (2:04:06.560)

of the JavaScript to machine code.

Lex Fridman (2:04:09.080)

For people who don't know, JIT is just in time compilation.

Travis Oliphant (2:04:11.520)

Thank you.

Lex Fridman (2:04:12.340)

Yeah, just in time compilation.

Travis Oliphant (2:04:13.240)

They're actually really sophisticated.

Lex Fridman (2:04:14.840)

In fact, I got jealous of how much effort

Travis Oliphant (2:04:17.100)

was put into the JavaScript JITs.

Lex Fridman (2:04:18.600)

Yes, well, it's kind of incredible what they've done.

Travis Oliphant (2:04:20.800)

Yes, I completely agree.

Lex Fridman (2:04:22.760)

I'm very impressed.

Lex Fridman (2:04:24.760)

But you know, Numba was an effort

Lex Fridman (2:04:26.880)

to make that happen with Python.

Lex Fridman (2:04:29.320)

And so we used some of the money

Lex Fridman (2:04:30.960)

we raised from Anaconda to do it.

Lex Fridman (2:04:32.440)

And then we also applied for this DARPA grant

Lex Fridman (2:04:34.800)

and used some of that money to continue the development.

Lex Fridman (2:04:36.820)

And then we used proceeds from service projects we would do.

Lex Fridman (2:04:40.680)

We get consulting projects

Travis Oliphant (2:04:41.800)

that we would then use some of the profits

Lex Fridman (2:04:44.480)

to invest in Numba.

Lex Fridman (2:04:45.400)

So we ended up with a team of two or three people

Lex Fridman (2:04:47.160)

working on Numba.

Lex Fridman (2:04:48.880)

It was a fits and starts, right?

Lex Fridman (2:04:50.720)

And ultimately, the fact that we had a commercial version

Travis Oliphant (2:04:53.560)

of it also we were writing.

Lex Fridman (2:04:54.720)

So part of the way I was trying to fund Numba,

Travis Oliphant (2:04:56.640)

say, well, let's do the free Numba

Lex Fridman (2:04:58.560)

and then we'll have a commercial version of Numba

Travis Oliphant (2:04:59.920)

called Numba Pro.

Lex Fridman (2:05:00.820)

And what Numba Pro did is it targeted GPUs.

Lex Fridman (2:05:03.240)

So we had the very first CUDA JIT

Lex Fridman (2:05:05.520)

and the very first at JIT compiler that in 2012 for 13,

Travis Oliphant (2:05:10.840)

you could run not just a view func on CPU,

Lex Fridman (2:05:14.140)

but a view func on GPUs.

Lex Fridman (2:05:15.640)

And it would automatically paralyze it

Lex Fridman (2:05:17.480)

and get 1000X speed on it.

Lex Fridman (2:05:18.840)

And that's an interesting funding mechanism

Lex Fridman (2:05:21.120)

because large companies or larger companies

Travis Oliphant (2:05:26.860)

care about speed in just this way.

Lex Fridman (2:05:30.120)

So it's exactly a really good way.

Travis Oliphant (2:05:33.140)

Yeah, there's been a couple of things

Lex Fridman (2:05:34.240)

you know people will pay for.

Lex Fridman (2:05:35.200)

One, they'll pay for really good user interfaces, right?

Lex Fridman (2:05:37.960)

And so I'm always looking for what are the things

Travis Oliphant (2:05:40.160)

people will pay for that you could actually adapt

Lex Fridman (2:05:41.720)

to the open source infrastructure?

Travis Oliphant (2:05:43.240)

One is definitely user interfaces.

Lex Fridman (2:05:45.560)

The second is speed, like a better runtime, faster runtime.

Lex Fridman (2:05:49.120)

And then when you say people,

Lex Fridman (2:05:50.000)

you mean like a small number of people pay a lot of money,

Lex Fridman (2:05:52.280)

but then there's also this other mechanism that.

Lex Fridman (2:05:54.440)

That's true.

Travis Oliphant (2:05:55.280)

A ton of people pay.

Lex Fridman (2:05:56.400)

That's true.

Travis Oliphant (2:05:57.220)

A little bit.

Lex Fridman (2:05:58.060)

First, I gotta, we mentioned Anaconda,

Travis Oliphant (2:06:00.320)

we mentioned friends, family, and fools.

Lex Fridman (2:06:04.280)

So Anaconda is yet another.

Lex Fridman (2:06:06.800)

So there's a company, but there's also a project.

Lex Fridman (2:06:09.080)

Correct.

Travis Oliphant (2:06:09.920)

That is exceptionally impactful in terms of,

Lex Fridman (2:06:14.600)

for many reasons, but one of which is bringing

Travis Oliphant (2:06:16.880)

a lot more people into the community

Lex Fridman (2:06:21.960)

of folks who use Python.

Lex Fridman (2:06:23.640)

So what is Anaconda?

Lex Fridman (2:06:26.920)

What is its goals?

Lex Fridman (2:06:28.960)

Maybe what is Conda versus Anaconda?

Lex Fridman (2:06:31.540)

Yeah, I'll tell you a little bit of the history of that.

Travis Oliphant (2:06:33.080)

Cause Anaconda, we wanted to do,

Lex Fridman (2:06:35.280)

we wanted to scale Python.

Travis Oliphant (2:06:37.440)

Cause we, you know, that was the goal.

Lex Fridman (2:06:38.680)

Peter and I had the goal of when we started Anaconda,

Travis Oliphant (2:06:40.720)

we actually started as Continuum Analytics

Lex Fridman (2:06:42.440)

was the name of the company that started.

Travis Oliphant (2:06:44.000)

It got renamed Anaconda in 2015.

Lex Fridman (2:06:47.000)

But we said, we want to scale analytics.

Travis Oliphant (2:06:49.880)

NumPy is great, Pandas is emerging,

Lex Fridman (2:06:52.680)

but these need to run at scale with lots of machines.

Travis Oliphant (2:06:55.320)

The other thing we wanted to do was make user interfaces

Lex Fridman (2:06:57.920)

that were web.

Travis Oliphant (2:06:59.360)

We wanted to make sure the web did not pass

Lex Fridman (2:07:01.320)

by the Python community.

Travis Oliphant (2:07:02.920)

That we had ways to translate your data science to the web.

Lex Fridman (2:07:06.000)

So those are the two kind of technical areas.

Travis Oliphant (2:07:07.720)

We thought, oh, we'll build products in this space.

Lex Fridman (2:07:09.920)

And that was the idea.

Travis Oliphant (2:07:12.500)

Very quickly in, but of course,

Lex Fridman (2:07:13.640)

the thing I knew how to do was to do consulting

Travis Oliphant (2:07:15.760)

to make money and to make sure my family and friends

Lex Fridman (2:07:18.920)

and fools that had invested didn't lose their money.

Lex Fridman (2:07:21.680)

So it's a little different

Lex Fridman (2:07:22.640)

than if you take money from a venture fund.

Travis Oliphant (2:07:24.360)

If you take money from a venture fund,

Lex Fridman (2:07:25.520)

the venture fund, they want you to go big or go home.

Lex Fridman (2:07:27.720)

And they're kind of like expecting nine out of 10 to fail

Lex Fridman (2:07:30.280)

or 99 out of 100 to fail.

Travis Oliphant (2:07:33.080)

It's different.

Lex Fridman (2:07:33.920)

I was, I was owed a barbell strategy.

Travis Oliphant (2:07:35.480)

I was like, I can't fail.

Lex Fridman (2:07:37.280)

I mean, I may not do super well,

Lex Fridman (2:07:38.680)

but I cannot lose their money.

Lex Fridman (2:07:40.440)

So I'm going to do something I know can return a profit,

Lex Fridman (2:07:43.560)

but I want to have exposure to an upside.

Lex Fridman (2:07:46.320)

So that's what happened at Anaconda.

Travis Oliphant (2:07:47.920)

We didn't, there was lots of things we did not well

Lex Fridman (2:07:50.320)

in terms of that structure.

Lex Fridman (2:07:51.320)

And I've learned from since and how to do it better.

Lex Fridman (2:07:53.740)

But we've, we did a really good job

Travis Oliphant (2:07:56.700)

of kind of attracting the interest around the area

Lex Fridman (2:07:59.140)

to get good people working

Lex Fridman (2:08:00.360)

and then get funnel some money

Lex Fridman (2:08:01.700)

on some interesting projects.

Travis Oliphant (2:08:03.080)

Super excited about what came out of our energy there.

Lex Fridman (2:08:05.200)

Like a lot did.

Lex Fridman (2:08:06.840)

So what are some of the interesting projects?

Lex Fridman (2:08:08.280)

So Dask, Numba, Bokeh, Conda.

Travis Oliphant (2:08:12.120)

There was a data shader, Panel, Holoviz.

Lex Fridman (2:08:16.200)

These are all tools that are extremely relevant

Travis Oliphant (2:08:19.040)

in terms of helping you build applications,

Lex Fridman (2:08:21.400)

build tools, build, you know, faster code.

Travis Oliphant (2:08:25.060)

There's a couple I'm forgetting.

Lex Fridman (2:08:25.900)

Oh, JupyterLab, JupyterLab came out of this too.

Lex Fridman (2:08:28.680)

And yeah.

Lex Fridman (2:08:30.320)

Okay, so Bokeh does plotting?

Lex Fridman (2:08:32.700)

Is that?

Lex Fridman (2:08:33.540)

Bokeh does plotting.

Lex Fridman (2:08:34.360)

So Bokeh was one of the foundational things to say,

Lex Fridman (2:08:35.880)

I want to do plot in Python,

Lex Fridman (2:08:37.360)

but have the things show up in a web.

Lex Fridman (2:08:39.140)

Right, that's right.

Travis Oliphant (2:08:40.040)

That's right, that's right.

Lex Fridman (2:08:40.880)

And plotting to me still,

Travis Oliphant (2:08:43.280)

with all due respect to Matplotlib and Bokeh,

Lex Fridman (2:08:46.480)

it feels like still an unsolved problem,

Travis Oliphant (2:08:48.760)

not a solved problem.

Lex Fridman (2:08:50.260)

It is, it's a big problem.

Travis Oliphant (2:08:52.160)

Right, because you're, I mean, I don't know,

Lex Fridman (2:08:55.640)

it's visualization broadly, right?

Travis Oliphant (2:08:58.640)

I think we've got a pretty good API story

Lex Fridman (2:09:00.960)

around certain use cases of plotting.

Lex Fridman (2:09:03.440)

But there's a difference between static plots

Lex Fridman (2:09:04.920)

versus interactive plots versus I'm an end user,

Travis Oliphant (2:09:07.800)

I just want to write a simple,

Lex Fridman (2:09:09.760)

for Pandas started the idea of here's a data frame

Travis Oliphant (2:09:12.040)

on a dot plot, I'm just going to attach plot

Lex Fridman (2:09:14.200)

as a method to my object,

Lex Fridman (2:09:16.380)

which was a little bit controversial, right?

Lex Fridman (2:09:18.280)

But works pretty well, actually,

Lex Fridman (2:09:20.160)

because there's a lot less you have to pass in, right?

Lex Fridman (2:09:23.680)

You can just say, here's my object, you know what you are,

Travis Oliphant (2:09:26.280)

you tell the visualization what to do.

Lex Fridman (2:09:29.000)

So that, and there's things like that

Travis Oliphant (2:09:31.320)

that have not been super well developed entirely,

Lex Fridman (2:09:33.720)

but Bokeh was focused on interactive plotting.

Lex Fridman (2:09:36.320)

So you could, it's a short path

Lex Fridman (2:09:38.400)

between interactive plotting and application,

Travis Oliphant (2:09:41.080)

dashboard application.

Lex Fridman (2:09:42.680)

And there's some incredible work that got done there, right?

Lex Fridman (2:09:44.760)

And it was a hard project,

Lex Fridman (2:09:45.800)

because then you're basically doing JavaScript and Python.

Lex Fridman (2:09:49.440)

So we wanted to tackle some of these hard problems

Lex Fridman (2:09:51.560)

and try to just go after them.

Travis Oliphant (2:09:53.440)

We got some DARPA funding to help,

Lex Fridman (2:09:54.920)

and it was super helpful, funny story there,

Travis Oliphant (2:09:56.880)

we actually did two DARPA proposals,

Lex Fridman (2:09:58.320)

but one we were five minutes late for.

Lex Fridman (2:10:00.580)

And DARPA has a very strict cutoff window.

Lex Fridman (2:10:03.040)

And so I, we had two proposals,

Travis Oliphant (2:10:04.760)

one for the Bokeh and one for actually Numba

Lex Fridman (2:10:06.720)

and the other work.

Lex Fridman (2:10:09.320)

Which one were you late for?

Lex Fridman (2:10:10.920)

The Foundation on Numerical Work.

Lex Fridman (2:10:12.920)

So Bokeh got funded. Oh no.

Lex Fridman (2:10:14.880)

Fortunately, Chris let us use some of the money to fund

Travis Oliphant (2:10:17.120)

still some of the other foundational work,

Lex Fridman (2:10:19.320)

but it wasn't as, yeah, his hands were tired,

Travis Oliphant (2:10:22.040)

he couldn't do anything about it.

Lex Fridman (2:10:23.880)

That was a whole interesting story.

Lex Fridman (2:10:25.880)

So one of the incredible projects

Lex Fridman (2:10:27.700)

that you worked on is Conda.

Travis Oliphant (2:10:29.200)

Yes.

Lex Fridman (2:10:30.040)

So what is Conda? So how that came about,

Travis Oliphant (2:10:31.400)

yeah, Conda, it was early on, like I said, with SciPy.

Lex Fridman (2:10:35.480)

SciPy was a distribution mass generation library.

Lex Fridman (2:10:37.880)

And he said, he heard me talking about compiler issues

Lex Fridman (2:10:40.320)

and trying to get the stuff shipped

Lex Fridman (2:10:41.480)

and the fact that people can use your libraries

Lex Fridman (2:10:43.320)

if they have it.

Lex Fridman (2:10:44.660)

So for a long time,

Lex Fridman (2:10:45.500)

we'd understood the packaging problem in Python.

Lex Fridman (2:10:47.800)

And one of the first things he did at Conda Analytics

Lex Fridman (2:10:50.680)

became Anaconda was organize the Pi data ecosystem

Travis Oliphant (2:10:54.240)

in conjunction with NumFocus.

Lex Fridman (2:10:56.160)

We actually started NumFocus

Travis Oliphant (2:10:58.960)

with some other folks in the community

Lex Fridman (2:11:00.480)

the same year we started Anaconda.

Travis Oliphant (2:11:02.880)

I said, we're gonna build a corporation,

Lex Fridman (2:11:04.200)

but we're also gonna reify the community aspect

Lex Fridman (2:11:07.040)

and build a nonprofit.

Lex Fridman (2:11:08.280)

So we did both of those.

Travis Oliphant (2:11:09.400)

Can we pause real quick and can you say what is PyPy,

Lex Fridman (2:11:13.280)

the Python package index,

Lex Fridman (2:11:14.720)

like this whole story of packaging in Python?

Lex Fridman (2:11:19.300)

Yeah, that's what I'm gonna get to actually.

Travis Oliphant (2:11:20.880)

This is exactly the journey I'm on.

Lex Fridman (2:11:22.240)

It's to sort of explain packaging in Python.

Travis Oliphant (2:11:24.200)

I think it's best expressed to the conversation

Lex Fridman (2:11:26.080)

I had with Guido at a conference,

Travis Oliphant (2:11:27.600)

where I said, so packaging is kind of a problem.

Lex Fridman (2:11:31.280)

And Guido said, I don't ever care about packaging.

Travis Oliphant (2:11:34.080)

I don't use it.

Lex Fridman (2:11:34.920)

I don't install new libraries.

Travis Oliphant (2:11:36.320)

I'm like, I guess if you're the language creator

Lex Fridman (2:11:38.200)

and if you need something, you just put it in the distribution

Travis Oliphant (2:11:40.480)

maybe you don't worry about packaging.

Lex Fridman (2:11:42.520)

But Guido has never really cared about packaging, right?

Lex Fridman (2:11:45.200)

And never really cared about the problem of distribution.

Lex Fridman (2:11:47.400)

It's somebody else's problem.

Lex Fridman (2:11:48.480)

And that's a fair position to take, I think,

Lex Fridman (2:11:50.240)

as a language creator.

Travis Oliphant (2:11:51.480)

In fact, there's a philosophical question about

Lex Fridman (2:11:54.160)

should you have different development packaging managers?

Lex Fridman (2:11:56.680)

Should you have a package manager per language?

Lex Fridman (2:11:58.400)

Is that really the right approach?

Travis Oliphant (2:11:59.800)

I think there are some answers of

Lex Fridman (2:12:01.900)

it is appropriate to have development tools.

Lex Fridman (2:12:04.200)

And there's an aspect of a development tool

Lex Fridman (2:12:06.040)

that is related to packaging.

Lex Fridman (2:12:07.680)

And every language should have some story there

Lex Fridman (2:12:10.600)

to help their developers create.

Lex Fridman (2:12:12.120)

So you should have language specific development tools.

Lex Fridman (2:12:14.960)

Development tools that relate to package managers.

Lex Fridman (2:12:17.080)

But then there's a very specific user story

Lex Fridman (2:12:19.520)

around package management

Travis Oliphant (2:12:20.680)

that those language specific package managers

Lex Fridman (2:12:22.240)

have to interact with.

Lex Fridman (2:12:23.560)

And currently aren't doing a good job of that.

Lex Fridman (2:12:25.920)

That was one of the challenges

Travis Oliphant (2:12:27.000)

that not seeing that difference,

Lex Fridman (2:12:29.140)

and it still exists in the difference today.

Travis Oliphant (2:12:31.720)

Conda always was a user.

Lex Fridman (2:12:34.480)

I'm gonna use Python to do data science.

Travis Oliphant (2:12:36.540)

I'm gonna use Python to do something.

Lex Fridman (2:12:38.240)

How do I get this installed?

Travis Oliphant (2:12:39.560)

It was always focused on that.

Lex Fridman (2:12:41.160)

So it didn't have a develop.

Travis Oliphant (2:12:43.880)

Classic example is pip has a pip develop.

Lex Fridman (2:12:45.960)

It's like, I wanna install this

Travis Oliphant (2:12:47.480)

into my current development environment today.

Lex Fridman (2:12:50.280)

Conda doesn't have that concept

Travis Oliphant (2:12:51.520)

because it's not part of the story.

Lex Fridman (2:12:52.840)

For people who don't know,

Travis Oliphant (2:12:54.640)

pip is a Python specific package manager.

Lex Fridman (2:12:59.640)

That's exceptionally popular.

Travis Oliphant (2:13:04.640)

That's probably like the default thing you've learned.

Lex Fridman (2:13:06.520)

It's the default user.

Lex Fridman (2:13:07.360)

And so the story there emerged

Lex Fridman (2:13:08.840)

because what happened is in 2012,

Travis Oliphant (2:13:11.480)

we had this meeting at the Googleplex

Lex Fridman (2:13:13.760)

and Guido was there to come talk about what we're gonna do,

Lex Fridman (2:13:15.600)

how we're gonna make things work better.

Lex Fridman (2:13:17.240)

And Wes McKinney, me, Peter,

Travis Oliphant (2:13:19.960)

Peter has a great photo of me talking to Guido

Lex Fridman (2:13:21.880)

and he pretends we're talking about this story.

Travis Oliphant (2:13:23.560)

Maybe we were, maybe we weren't.

Lex Fridman (2:13:24.680)

But we did at that meeting talk about it

Lex Fridman (2:13:26.320)

and asked Guido, we need to fix packaging in Python.

Lex Fridman (2:13:29.920)

People can't get the stuff.

Lex Fridman (2:13:31.040)

And he said, go fix it yourself.

Lex Fridman (2:13:32.400)

I don't think we're gonna do it.

Travis Oliphant (2:13:33.600)

All right.

Lex Fridman (2:13:35.720)

The origin story right there.

Travis Oliphant (2:13:36.960)

All right, you said, okay, you said to do this ourselves.

Lex Fridman (2:13:39.440)

So at the same time,

Travis Oliphant (2:13:41.640)

people did start to work on the packaging story in Python.

Lex Fridman (2:13:44.600)

It just took a little longer.

Lex Fridman (2:13:45.680)

So in 2012, kind of motivated

Lex Fridman (2:13:48.160)

by our training courses we were teaching,

Travis Oliphant (2:13:49.600)

like very similar to what you just mentioned

Lex Fridman (2:13:51.440)

about your mother.

Travis Oliphant (2:13:52.280)

Like it was motivated by the same purpose.

Lex Fridman (2:13:54.160)

Like how do we get this into people's hands?

Travis Oliphant (2:13:56.040)

It's this big, long process.

Lex Fridman (2:13:57.120)

It takes too expensive.

Travis Oliphant (2:13:58.520)

It was actually hurting NumPy development

Lex Fridman (2:14:00.200)

because I would hear people were saying,

Travis Oliphant (2:14:02.280)

don't make that change to NumPy

Lex Fridman (2:14:03.480)

because I just spent a week getting my Python environment.

Lex Fridman (2:14:05.480)

And if you change NumPy, I have to reinstall everything.

Lex Fridman (2:14:09.160)

And reinstalling is such a pain, don't do it.

Travis Oliphant (2:14:10.880)

I'm like, wait, okay.

Lex Fridman (2:14:12.120)

So now we're not making changes to a library

Travis Oliphant (2:14:14.640)

because of the installation problem

Lex Fridman (2:14:16.000)

that it'll cause for end users.

Travis Oliphant (2:14:17.440)

Okay, there's a problem with installation.

Lex Fridman (2:14:19.400)

We gotta fix this.

Lex Fridman (2:14:20.520)

So we said, we're gonna make a distribution in Python.

Lex Fridman (2:14:23.760)

And we'd previously done that.

Travis Oliphant (2:14:24.760)

I'd previously done that at mthought.

Lex Fridman (2:14:26.920)

I wanted to make one that would give away for free,

Travis Oliphant (2:14:28.520)

that everyone could just get.

Lex Fridman (2:14:29.840)

Like that was critical that we could just get it.

Travis Oliphant (2:14:32.080)

It wasn't tied to a product.

Lex Fridman (2:14:33.880)

It was just you could get it.

Lex Fridman (2:14:35.360)

And then we had constantly thought about,

Lex Fridman (2:14:36.960)

well, do we just leverage RPM?

Lex Fridman (2:14:39.120)

But the challenge had always been,

Lex Fridman (2:14:40.400)

we want a package manager that works on Windows,

Lex Fridman (2:14:42.240)

Mac OS X, and Linux the same, right?

Lex Fridman (2:14:45.040)

And it wasn't there.

Travis Oliphant (2:14:46.560)

Like you don't have anything like that.

Lex Fridman (2:14:47.960)

You have...

Lex Fridman (2:14:48.800)

And for people who don't know,

Lex Fridman (2:14:49.640)

RPM is an operating system specific package manager.

Travis Oliphant (2:14:54.560)

Correct, it's an operating specific.

Lex Fridman (2:14:55.960)

Yes, exactly.

Lex Fridman (2:14:56.800)

So do you create the design questions,

Lex Fridman (2:15:00.160)

do you create an umbrella package manager

Lex Fridman (2:15:02.240)

that works across operating systems?

Lex Fridman (2:15:03.840)

Yes, that was the decision.

Lex Fridman (2:15:05.680)

And in neighboring design questions,

Lex Fridman (2:15:08.080)

do you also create a package manager

Lex Fridman (2:15:09.920)

that spans multiple programming languages?

Lex Fridman (2:15:11.840)

Correct, exactly.

Travis Oliphant (2:15:12.760)

That was the world we faced.

Lex Fridman (2:15:14.280)

And we decided to go multiple operating systems,

Travis Oliphant (2:15:17.080)

multiple and programming language independent.

Lex Fridman (2:15:19.220)

Because even Python, and particularly what was important

Lex Fridman (2:15:21.800)

was SciPy has a bunch of Fortran in it, right?

Lex Fridman (2:15:24.920)

And scikit learn has links to a bunch of C++.

Travis Oliphant (2:15:27.760)

There's a lot of compiled code.

Lex Fridman (2:15:29.960)

And the Python package managers, especially early on,

Travis Oliphant (2:15:32.920)

didn't even support that.

Lex Fridman (2:15:34.320)

So in 2000, so we released Anaconda,

Travis Oliphant (2:15:38.520)

which was just a distribution of libraries,

Lex Fridman (2:15:39.960)

but we started to work on Conda in 2012.

Travis Oliphant (2:15:42.480)

First version of Conda came out in early 2013,

Lex Fridman (2:15:44.680)

summer of 2013, and it was a package manager.

Lex Fridman (2:15:47.840)

So you could say, Conda install scikit learn.

Lex Fridman (2:15:49.560)

In fact, scikit learn was a fantastic project that emerged.

Travis Oliphant (2:15:54.280)

It was the classic example of the scikits.

Lex Fridman (2:15:57.120)

I talked to you earlier about SciPy being too big

Travis Oliphant (2:15:59.760)

to be a single library.

Lex Fridman (2:16:01.240)

Well, what the community had done is said,

Travis Oliphant (2:16:02.680)

let's make scikits.

Lex Fridman (2:16:04.160)

And there's scikit image, there's scikit learn,

Travis Oliphant (2:16:05.840)

there's a lot of scikits.

Lex Fridman (2:16:07.640)

And it was a fantastic move that the community did.

Travis Oliphant (2:16:10.200)

I didn't do it.

Lex Fridman (2:16:11.460)

I was like, okay, that's a good idea.

Travis Oliphant (2:16:12.560)

I didn't like the name.

Lex Fridman (2:16:13.540)

I didn't like the fact you typed scikit image.

Travis Oliphant (2:16:15.500)

I was like, that's gotta be simpler.

Lex Fridman (2:16:17.400)

That's scikit learn, we gotta make that smaller.

Travis Oliphant (2:16:19.800)

I don't like typing all this stuff from imports.

Lex Fridman (2:16:21.940)

So I was kind of a pressure that way,

Lex Fridman (2:16:23.220)

but I love the energy and love the fact

Lex Fridman (2:16:25.280)

that they went out and they did it,

Lex Fridman (2:16:26.200)

and DOS people, Jared Millman, and then of course, Gael,

Lex Fridman (2:16:29.400)

and there's people I'm not even naming.

Travis Oliphant (2:16:31.280)

Scikit learn really emerged as a fantastic project.

Lex Fridman (2:16:34.640)

And the documentation around that is also incredible.

Lex Fridman (2:16:36.680)

And the documentation was incredible, exactly.

Lex Fridman (2:16:37.840)

I don't know who did that, but they did a great job.

Travis Oliphant (2:16:40.160)

A lot of people in Inria, a lot of European contributors.

Lex Fridman (2:16:45.120)

There's some Andreas in the US.

Travis Oliphant (2:16:47.120)

There's a lot of just people I just adore,

Lex Fridman (2:16:48.920)

I think are amazing people.

Lex Fridman (2:16:51.180)

Awesome use of SciPy, right?

Lex Fridman (2:16:52.480)

I love the fact that they were using SciPy effectively

Travis Oliphant (2:16:54.600)

to do something I love, which is machine learning,

Lex Fridman (2:16:57.160)

but couldn't install it.

Travis Oliphant (2:16:58.980)

Because there's so many pieces involved.

Lex Fridman (2:17:00.600)

So many dependencies, right?

Lex Fridman (2:17:02.160)

So our use case of Conda was Conda install scikit learn.

Lex Fridman (2:17:06.080)

Right, and it was the best way to install scikit learn

Travis Oliphant (2:17:09.440)

in 2013 to really 2018, 17, 18, PIP finally caught up.

Lex Fridman (2:17:14.440)

I still think it's you should Conda install scikit learn

Travis Oliphant (2:17:16.440)

for the PIP install scikit learn,

Lex Fridman (2:17:17.560)

but you can PIP install scikit learn.

Travis Oliphant (2:17:19.360)

The issue is the package they created was wheels

Lex Fridman (2:17:21.840)

and PIP does not handle the multi vendor approach.

Travis Oliphant (2:17:24.480)

They don't handle the fact you have C++ libraries

Lex Fridman (2:17:26.600)

you're depending on.

Travis Oliphant (2:17:27.680)

They just stop at the Python boundary.

Lex Fridman (2:17:29.240)

And so what you have to do in the wheel world

Travis Oliphant (2:17:31.280)

is you have to vendor.

Lex Fridman (2:17:33.200)

You have to take all of the binary and vendor it.

Travis Oliphant (2:17:35.640)

Now, if your change happens in underlying dependency,

Lex Fridman (2:17:38.480)

you have to redo the whole wheel.

Lex Fridman (2:17:40.280)

So TensorFlow, as you know,

Lex Fridman (2:17:42.080)

you should not PIP install TensorFlow.

Travis Oliphant (2:17:44.680)

It's a terrible idea.

Lex Fridman (2:17:45.520)

People do it because the popularity of PIP,

Travis Oliphant (2:17:48.640)

many people think, oh, of course,

Lex Fridman (2:17:49.480)

that's how I install everything in Python.

Travis Oliphant (2:17:51.480)

Yeah, this is one of the big challenges.

Lex Fridman (2:17:53.960)

You take a GitHub repository or just a basic blog post.

Travis Oliphant (2:17:57.920)

The number of time PIP is mentioned over Conda

Lex Fridman (2:18:00.840)

is like 100 X to one.

Travis Oliphant (2:18:02.760)

Correct, correct.

Lex Fridman (2:18:03.600)

So it just has to do with the.

Lex Fridman (2:18:04.440)

And that was increasing.

Lex Fridman (2:18:05.280)

It wasn't true early because PIP didn't exist.

Travis Oliphant (2:18:07.520)

Like Conda came first.

Lex Fridman (2:18:08.840)

So but that's the problem.

Travis Oliphant (2:18:10.160)

Like Conda came first, but that's like the long tail

Lex Fridman (2:18:13.040)

of the internet documentation user generated.

Lex Fridman (2:18:15.840)

So that like you think, how do I install Google?

Lex Fridman (2:18:19.160)

How do I install TensorFlow?

Travis Oliphant (2:18:20.400)

You're just not gonna see Conda in that first page.

Lex Fridman (2:18:23.000)

Correct, exactly.

Lex Fridman (2:18:24.120)

And that.

Lex Fridman (2:18:24.960)

Not today, you would have in 2016, 2017.

Lex Fridman (2:18:29.400)

And it's sad because Conda solves

Lex Fridman (2:18:32.760)

a lot of usability issues.

Travis Oliphant (2:18:34.160)

Correct.

Lex Fridman (2:18:35.000)

Like for especially super challenging thing.

Travis Oliphant (2:18:36.480)

I don't know.

Lex Fridman (2:18:37.320)

One of the big pain points for me was

Travis Oliphant (2:18:39.520)

just on the computer vision side, OpenCV installation.

Lex Fridman (2:18:43.560)

Perfect example.

Travis Oliphant (2:18:44.400)

Yes.

Lex Fridman (2:18:45.240)

I think Conda, I don't know if Conda solved that one.

Travis Oliphant (2:18:47.400)

Conda has an OpenCV package.

Lex Fridman (2:18:49.080)

I don't know.

Travis Oliphant (2:18:49.920)

I certainly know PIP has not solved.

Lex Fridman (2:18:53.440)

I mean, there's complexities there because.

Travis Oliphant (2:18:55.840)

Right.

Lex Fridman (2:18:56.680)

I actually don't know.

Travis Oliphant (2:18:57.640)

I should probably know a good answer for this,

Lex Fridman (2:18:59.120)

but if you compile OpenCV with certain dependencies,

Travis Oliphant (2:19:05.440)

you'll be able to do certain things.

Lex Fridman (2:19:07.440)

So there's this kind of flexibility of what you,

Travis Oliphant (2:19:09.840)

like what options you compile with.

Lex Fridman (2:19:12.960)

Yes.

Lex Fridman (2:19:13.800)

And I don't think it's trivial to do that with Conda or.

Lex Fridman (2:19:17.840)

So Conda has a notion of variance of a package.

Travis Oliphant (2:19:20.520)

You can actually have different compilation versions

Lex Fridman (2:19:23.120)

of a package.

Lex Fridman (2:19:23.960)

So not just the version is different,

Lex Fridman (2:19:24.800)

but oh, this is compiled with these optimizations on.

Lex Fridman (2:19:26.880)

So Conda does have an answer.

Lex Fridman (2:19:28.000)

Has those flavors.

Travis Oliphant (2:19:28.840)

Has flavors, basically.

Lex Fridman (2:19:30.080)

Well, PIP, as far as I know, does not have flavors.

Travis Oliphant (2:19:32.360)

No, no.

Lex Fridman (2:19:33.280)

PIP generally hasn't thought deeply

Lex Fridman (2:19:36.440)

about the binary dependency problem, right?

Lex Fridman (2:19:38.400)

And that's why fundamentally it doesn't work

Travis Oliphant (2:19:41.840)

for the SciPy ecosystem.

Lex Fridman (2:19:43.640)

It barely, you can sort of paper over it and duct tape

Lex Fridman (2:19:46.120)

and it kind of works until it doesn't

Lex Fridman (2:19:48.040)

and it falls apart entirely.

Lex Fridman (2:19:49.560)

So it's been a mixed bag.

Lex Fridman (2:19:51.520)

Like, and I've been having lots of conversations

Travis Oliphant (2:19:54.360)

with people over the years because again,

Lex Fridman (2:19:56.120)

it's an area where if you understand some things,

Lex Fridman (2:19:58.400)

but not all the things,

Lex Fridman (2:19:59.240)

but they've done a great job of community appeal.

Travis Oliphant (2:20:02.200)

This is an area where I think Anaconda as a company

Lex Fridman (2:20:05.560)

needed to do some things

Lex Fridman (2:20:07.040)

in order to make Conda more community centric, right?

Lex Fridman (2:20:10.440)

And this is a, I talk about this all the time.

Travis Oliphant (2:20:13.080)

There's a balance between you have every project starts

Lex Fridman (2:20:16.640)

with what I called company backed open source.

Travis Oliphant (2:20:18.280)

Even if the company is yourself, it's just one person,

Lex Fridman (2:20:20.320)

just doing business as.

Lex Fridman (2:20:23.360)

But ultimately for products to succeed virally

Lex Fridman (2:20:26.080)

and become massive influencers,

Travis Oliphant (2:20:28.320)

they have to create,

Lex Fridman (2:20:29.160)

they have to get community people on board.

Travis Oliphant (2:20:30.520)

They have to get other people on board.

Lex Fridman (2:20:32.120)

So it has to become community driven.

Lex Fridman (2:20:33.680)

And a big part of that is engagement with those people.

Lex Fridman (2:20:35.520)

Empowering people, governance around it.

Lex Fridman (2:20:38.600)

And what happened with Conda in the early days,

Lex Fridman (2:20:41.360)

PIP emerged and we did do some good things.

Travis Oliphant (2:20:43.720)

Conda Forge, Conda Forge community

Lex Fridman (2:20:46.400)

is sort of the community recipe creation community.

Lex Fridman (2:20:49.880)

But Conda itself, I still believe,

Lex Fridman (2:20:52.160)

and Peter is CEO of Anaconda, he's my co founder.

Travis Oliphant (2:20:55.120)

I ran Anaconda until 2017, 2018.

Lex Fridman (2:20:58.160)

Is Peter still Anaconda?

Lex Fridman (2:20:59.000)

Peter's still Anaconda, right?

Lex Fridman (2:21:00.000)

And we're still great friends.

Travis Oliphant (2:21:01.360)

We talk all the time.

Lex Fridman (2:21:02.560)

I love him to death.

Travis Oliphant (2:21:03.600)

There's a long story there about like why and how

Lex Fridman (2:21:06.080)

and we can cover in some other podcast perhaps.

Travis Oliphant (2:21:08.640)

Yeah.

Lex Fridman (2:21:09.480)

It's sort of a more, maybe a more business focused one.

Lex Fridman (2:21:11.400)

But this is one area where I think Conda

Lex Fridman (2:21:15.160)

should be more community driven.

Travis Oliphant (2:21:17.280)

Like he should be pushing more

Lex Fridman (2:21:18.960)

to get more community contributors to Conda

Lex Fridman (2:21:21.200)

and let the, Anaconda shouldn't be fighting this battle.

Lex Fridman (2:21:26.080)

Yeah.

Lex Fridman (2:21:26.920)

Right?

Lex Fridman (2:21:27.760)

It's actually, it's really a developers.

Travis Oliphant (2:21:28.600)

Like you said, like help the developers

Lex Fridman (2:21:30.400)

and then they'll actually move us the right direction.

Travis Oliphant (2:21:32.200)

Well, that was the problem I have is many

Lex Fridman (2:21:34.040)

of the cool kids I know don't use Conda.

Lex Fridman (2:21:36.520)

And that to me is confusing.

Lex Fridman (2:21:38.880)

It is confusing.

Travis Oliphant (2:21:39.800)

It's really a matter of, Conda has some challenges.

Lex Fridman (2:21:42.640)

First of all, Conda still needs to be improved.

Travis Oliphant (2:21:44.120)

There's lots of improvements to be made.

Lex Fridman (2:21:45.320)

And it's that aspect of wait, who's doing this?

Lex Fridman (2:21:47.600)

And the fact that then the Pi PA really stepped up.

Lex Fridman (2:21:50.960)

Like they were not solving the problem at all.

Lex Fridman (2:21:53.400)

And now they kind of got to where they're solving it

Lex Fridman (2:21:55.640)

for the most part.

Lex Fridman (2:21:56.720)

And then effectively you could get,

Lex Fridman (2:21:58.160)

like Conda solved a problem that was there.

Lex Fridman (2:22:00.360)

And it still does.

Lex Fridman (2:22:01.200)

It's still, you know, there's still great things it can do.

Travis Oliphant (2:22:03.960)

But, and we still use it all the time at one site

Lex Fridman (2:22:06.920)

and with other clients, but with,

Lex Fridman (2:22:08.960)

but you can kind of do similar things with PIP and Docker.

Lex Fridman (2:22:12.160)

Right?

Lex Fridman (2:22:13.000)

So especially with the web development community,

Lex Fridman (2:22:15.280)

that part of it, again, is this is the,

Travis Oliphant (2:22:17.080)

there's a lot of different kinds of developers

Lex Fridman (2:22:19.200)

in the Python ecosystem.

Lex Fridman (2:22:20.200)

And there's still a lack of some clear understanding.

Lex Fridman (2:22:23.720)

I go to the Python conference all the time

Lex Fridman (2:22:25.320)

and then there's only a few people in the Pi PA who get it.

Lex Fridman (2:22:28.280)

And then others who are just massively trumpeting

Travis Oliphant (2:22:30.680)

the power of PIP, but just do not understand the problem.

Lex Fridman (2:22:32.840)

Yeah.

Lex Fridman (2:22:33.680)

So one of the obvious things to me from a mom,

Lex Fridman (2:22:36.040)

from a non programmer perspective,

Travis Oliphant (2:22:37.840)

is the across operating system usability.

Lex Fridman (2:22:41.760)

That's much more natural.

Lex Fridman (2:22:42.680)

So there's people that use Windows and just,

Lex Fridman (2:22:45.440)

it seems much easier to recommend Conda there,

Lex Fridman (2:22:49.080)

but then it, you should also recommend it across the board.

Lex Fridman (2:22:51.840)

So I'll definitely sort of.

Lex Fridman (2:22:53.520)

But what I recommend now is a hybrid.

Lex Fridman (2:22:55.320)

I do.

Travis Oliphant (2:22:56.160)

I mean, I have no problem.

Lex Fridman (2:22:57.000)

Is it possible to use?

Travis Oliphant (2:22:57.840)

Oh, it is.

Lex Fridman (2:22:58.660)

It is.

Lex Fridman (2:22:59.500)

But like build the environment with PIP, with Conda,

Lex Fridman (2:23:01.600)

build an environment with Conda

Lex Fridman (2:23:03.360)

and then PIP install on top of that.

Lex Fridman (2:23:04.600)

That's fine.

Travis Oliphant (2:23:05.440)

Be careful about PIP installing OpenCV or TensorFlow

Lex Fridman (2:23:09.400)

or because if somebody's allowed that,

Travis Oliphant (2:23:11.360)

it's gonna be most surely done in a way

Lex Fridman (2:23:13.320)

that can't be updated that easily.

Lex Fridman (2:23:15.120)

So install like the big packages,

Lex Fridman (2:23:17.680)

the infrastructure with Conda and then the weirdos.

Travis Oliphant (2:23:21.000)

Yeah.

Lex Fridman (2:23:21.840)

That like the weird like implementation for some.

Travis Oliphant (2:23:24.720)

I had a, there's a cool library I used

Lex Fridman (2:23:28.440)

that based on your location and time of day and date

Travis Oliphant (2:23:33.520)

tells you the exact position of the sun

Lex Fridman (2:23:35.640)

relative to the earth.

Lex Fridman (2:23:38.160)

And it's just like a simple library,

Lex Fridman (2:23:39.700)

but it's very precise.

Lex Fridman (2:23:41.360)

And I was like, all right.

Lex Fridman (2:23:42.200)

But that was, that was, and it's like PIP.

Travis Oliphant (2:23:45.120)

Well, the thing they did really well is Python developers

Lex Fridman (2:23:48.600)

who wanna get their stuff published,

Travis Oliphant (2:23:50.600)

you have to have a PIP recipe.

Lex Fridman (2:23:51.920)

Yeah.

Lex Fridman (2:23:52.760)

Right?

Lex Fridman (2:23:53.600)

I mean, even if it's, you know, the challenge is,

Lex Fridman (2:23:56.440)

and there's a key thing that needs to be added to PIP,

Lex Fridman (2:23:58.800)

just simply add to PIP the ability to defer

Travis Oliphant (2:24:01.680)

to a system package manager.

Lex Fridman (2:24:03.440)

Like, cause it's, you know,

Travis Oliphant (2:24:04.460)

recognize you're not gonna solve all the dependency problem.

Lex Fridman (2:24:07.280)

So let like give up and allow the system package to work.

Travis Oliphant (2:24:12.420)

That way Anaconda is installed and it has PIP.

Lex Fridman (2:24:15.140)

It would default to Conda to install stuff,

Lex Fridman (2:24:16.960)

but Red Hat RPM would default to RPM

Lex Fridman (2:24:19.240)

to install some more things.

Travis Oliphant (2:24:20.600)

Like that's the, that's a key, not difficult,

Lex Fridman (2:24:23.480)

but somewhat work, some work feature needs to be added.

Travis Oliphant (2:24:25.960)

That's an example of something like,

Lex Fridman (2:24:27.440)

I've known we need to do it.

Travis Oliphant (2:24:28.620)

I mean, it's where I wish I had more money.

Lex Fridman (2:24:30.920)

I wish I was more successful in the business side,

Travis Oliphant (2:24:33.480)

trying to get there, but I wish my, you know,

Lex Fridman (2:24:35.060)

my family, friends and full community that I know.

Travis Oliphant (2:24:37.280)

Was larger.

Lex Fridman (2:24:38.120)

Was larger and had more money.

Travis Oliphant (2:24:39.320)

Cause I know tons of things to do effectively

Lex Fridman (2:24:42.680)

with more resources, but you know,

Travis Oliphant (2:24:46.280)

I have not yet been successful at channel.

Lex Fridman (2:24:48.720)

Tons of, you know, some, you know,

Travis Oliphant (2:24:49.960)

I'm happy with what we've done.

Lex Fridman (2:24:51.480)

We created again at Quansight,

Lex Fridman (2:24:54.840)

what we created to get Anaconda started.

Lex Fridman (2:24:56.480)

We created community to get Anaconda started.

Travis Oliphant (2:24:58.160)

Done it again with Quansight.

Lex Fridman (2:24:59.280)

Super excited by that.

Lex Fridman (2:25:00.480)

But it took three years to do it.

Lex Fridman (2:25:02.200)

What is Quansight?

Lex Fridman (2:25:03.200)

What is its mission?

Lex Fridman (2:25:04.440)

We've talked a few times about different fascinating

Travis Oliphant (2:25:06.920)

aspects of it, but let's like big picture,

Lex Fridman (2:25:08.920)

what is Quansight?

Travis Oliphant (2:25:09.760)

Big picture Quansight.

Lex Fridman (2:25:10.600)

Quansight is, its mission is to connect data

Travis Oliphant (2:25:13.480)

to an open economy.

Lex Fridman (2:25:14.480)

So it's basically consulting of the pie data ecosystem,

Lex Fridman (2:25:17.520)

right?

Lex Fridman (2:25:18.360)

It's a consulting company.

Lex Fridman (2:25:19.280)

And what I've said when I started it was we're trying

Lex Fridman (2:25:21.200)

to create products, people, and technology.

Lex Fridman (2:25:24.700)

So it's divided into two groups.

Lex Fridman (2:25:26.700)

And a third one as well.

Travis Oliphant (2:25:28.300)

The two groups are a consulting services company

Lex Fridman (2:25:30.360)

that just helps people do data science

Lex Fridman (2:25:31.960)

and data engineering and data management better

Lex Fridman (2:25:35.080)

and more efficiently.

Travis Oliphant (2:25:35.920)

Like full stack, like full thing.

Lex Fridman (2:25:36.760)

Full stack data science, full thing.

Travis Oliphant (2:25:38.200)

We'll help you build a infrastructure.

Lex Fridman (2:25:40.020)

If you're using Jupiter, we need,

Travis Oliphant (2:25:41.380)

we do staff augmentation, need more pro programmers,

Lex Fridman (2:25:43.820)

help you use Dask more effectively,

Travis Oliphant (2:25:44.900)

help you use GPUs more effectively.

Lex Fridman (2:25:46.520)

Just basically a lot of people need help.

Lex Fridman (2:25:48.400)

So we do training as well to help people, you know,

Lex Fridman (2:25:50.800)

both immediate help and then get, learn from somebody.

Travis Oliphant (2:25:55.860)

We've added a bunch of stuff too.

Lex Fridman (2:25:57.080)

We've kind of separated some of these other things

Travis Oliphant (2:25:58.600)

into another company called Open Teams

Lex Fridman (2:26:00.120)

that we currently started.

Travis Oliphant (2:26:01.760)

One of the things I loved about what we did at Anaconda

Lex Fridman (2:26:03.380)

was creating a community innovation team.

Lex Fridman (2:26:05.520)

And so I wanted to replicate that.

Lex Fridman (2:26:06.700)

This time we did a lot of innovation at Anaconda.

Travis Oliphant (2:26:09.360)

I wanted to do innovation,

Lex Fridman (2:26:10.600)

but also contribute to the projects that existed,

Travis Oliphant (2:26:13.680)

like create a place where maintainers,

Lex Fridman (2:26:16.440)

so the SciPy and NumPy and Numba

Lex Fridman (2:26:18.480)

and all these projects we already started

Lex Fridman (2:26:20.400)

can pay people to work on them and keep them going.

Lex Fridman (2:26:22.700)

So that's Labs.

Lex Fridman (2:26:23.540)

Quansight Labs is a separate organization.

Travis Oliphant (2:26:25.960)

It's a nonprofit mission.

Lex Fridman (2:26:28.060)

The profits of Quansight help fund it.

Lex Fridman (2:26:29.940)

And in fact, every project that we have at Quansight,

Lex Fridman (2:26:33.240)

a portion of the money goes directly to Quansight Labs

Travis Oliphant (2:26:36.040)

to help keep it funded.

Lex Fridman (2:26:37.060)

So we've gotten several mechanisms

Travis Oliphant (2:26:38.280)

that we keep Quansight Labs funded.

Lex Fridman (2:26:40.040)

And currently, so I'm really excited about Labs

Travis Oliphant (2:26:41.960)

because it's been a mission for a long time.

Lex Fridman (2:26:43.680)

What kind of projects are within Labs?

Lex Fridman (2:26:45.240)

So Labs is working to make the software better,

Lex Fridman (2:26:47.680)

like make NumPy better, make SciPy better.

Travis Oliphant (2:26:49.760)

It only works on open source.

Lex Fridman (2:26:52.340)

So if somebody wants to, so companies do,

Travis Oliphant (2:26:55.440)

we have a thing called a community work order, we call it.

Lex Fridman (2:26:57.480)

If a company says, I wanna make Spyder better.

Travis Oliphant (2:27:00.020)

Okay, cool.

Lex Fridman (2:27:01.680)

You can pay for a month of a developer of Spyder

Travis Oliphant (2:27:05.440)

or a developer of NumPy or a developer of SciPy.

Lex Fridman (2:27:08.400)

You can't tell them what you want them to do.

Travis Oliphant (2:27:09.840)

You can give them your priorities and things you wish existed

Lex Fridman (2:27:12.800)

and they'll work on those priorities with the community

Travis Oliphant (2:27:16.080)

to get what the community wants

Lex Fridman (2:27:17.560)

and what emerges of what the community wants.

Travis Oliphant (2:27:18.880)

Is there some aspect on the consulting side

Lex Fridman (2:27:21.080)

that is helping, as we were talking about morphology

Lex Fridman (2:27:24.320)

and so on, is there specific application

Lex Fridman (2:27:26.600)

that are particularly like driving,

Lex Fridman (2:27:29.120)

sort of inspiring the need for updates to SciPy?

Lex Fridman (2:27:32.000)

Correct, absolutely, absolutely.

Travis Oliphant (2:27:33.360)

GPUs are absolutely one of them.

Lex Fridman (2:27:34.840)

And new hardware beyond GPUs.

Travis Oliphant (2:27:36.840)

I mean, Tesla's Dojo chip, I'm hoping we'll have a chance

Lex Fridman (2:27:39.720)

to work on that perhaps.

Travis Oliphant (2:27:42.320)

Things like that are definitely driving it.

Lex Fridman (2:27:43.840)

The other thing that's driving it is scalable,

Travis Oliphant (2:27:45.520)

like speed and scale.

Lex Fridman (2:27:47.640)

How do I write NumPy code or NumPy Lite code

Lex Fridman (2:27:50.360)

if I want it to run across a cluster?

Lex Fridman (2:27:52.520)

That's Dask or maybe it's Ray.

Travis Oliphant (2:27:54.240)

I mean, there's sort of ways to do that now.

Lex Fridman (2:27:56.360)

Or there's Moden and there's, so Pandas code,

Travis Oliphant (2:27:59.720)

NumPy code, SciPy code, Scikit learn code

Lex Fridman (2:28:02.080)

that I want to scale.

Lex Fridman (2:28:03.240)

So that's one big area.

Lex Fridman (2:28:04.880)

Have you gotten a chance to chat with Andre and Elon

Travis Oliphant (2:28:08.400)

about particular, because like.

Lex Fridman (2:28:09.840)

No, I would love to, by the way.

Travis Oliphant (2:28:11.360)

I have not, but I'd love to.

Lex Fridman (2:28:12.280)

I just saw their Tesla AI Days video.

Travis Oliphant (2:28:15.520)

Super excited.

Lex Fridman (2:28:16.360)

That's one of the, you know, I love great engineering,

Travis Oliphant (2:28:18.600)

software engineering teams and engineering teams in general.

Lex Fridman (2:28:21.000)

And they're doing a lot of incredible stuff with Python.

Travis Oliphant (2:28:23.040)

They're like revolutionary.

Lex Fridman (2:28:25.040)

So many aspects of the machine learning pipeline.

Travis Oliphant (2:28:28.800)

I agree.

Lex Fridman (2:28:29.640)

That's operating in the real world.

Lex Fridman (2:28:30.600)

And so much of that is Python.

Lex Fridman (2:28:31.880)

Like you said, the guy running, you know, Andre Kapathy,

Travis Oliphant (2:28:35.000)

running Autopilot is tweeting about optimization

Lex Fridman (2:28:38.680)

of NumPy versus.

Travis Oliphant (2:28:41.200)

I would love to talk to him.

Lex Fridman (2:28:42.920)

In fact, we have at Quonset, we've been fortunate enough

Travis Oliphant (2:28:45.080)

to work with Facebook on PyTorch directly.

Lex Fridman (2:28:47.560)

So we have about 13 developers at Quonset.

Travis Oliphant (2:28:49.880)

Some of them are in labs working directly on PyTorch.

Lex Fridman (2:28:52.560)

On PyTorch.

Travis Oliphant (2:28:53.400)

On PyTorch, right.

Lex Fridman (2:28:54.240)

So I basically started Quonset.

Travis Oliphant (2:28:55.680)

I went to both TensorFlow and PyTorch and said,

Lex Fridman (2:28:57.160)

hey, I want to help connect what you're doing

Travis Oliphant (2:29:00.200)

to the broader SciPy ecosystem.

Lex Fridman (2:29:01.920)

Because I see what you're doing.

Travis Oliphant (2:29:03.240)

We have this bigger mission that we want to make sure

Lex Fridman (2:29:04.760)

we don't, you know, lose energy here.

Travis Oliphant (2:29:06.760)

So, and Facebook responded really positively

Lex Fridman (2:29:09.840)

and I didn't get the same reaction.

Travis Oliphant (2:29:12.400)

Not yet, not yet.

Lex Fridman (2:29:13.560)

Not yet.

Lex Fridman (2:29:14.400)

So I really love the folks at TensorFlow, too.

Lex Fridman (2:29:17.480)

They're fantastic.

Travis Oliphant (2:29:18.480)

I think it's the, just how it integrates

Lex Fridman (2:29:21.120)

with their business.

Travis Oliphant (2:29:21.960)

I mean, like I said, there's a lot of reasons.

Lex Fridman (2:29:23.800)

Just the timing, the integration with their business,

Lex Fridman (2:29:25.720)

what they're looking for.

Lex Fridman (2:29:27.160)

They're probably looking for more users.

Lex Fridman (2:29:28.760)

And I was looking to kind of cut up some development effort

Lex Fridman (2:29:31.600)

and they couldn't receive that as easily, I think.

Lex Fridman (2:29:33.840)

So I'm hoping, I'm really hopeful

Lex Fridman (2:29:36.040)

and love the people there.

Lex Fridman (2:29:37.640)

What's the idea behind OpenTeams?

Lex Fridman (2:29:39.800)

So OpenTeams, I'm super excited about OpenTeams

Travis Oliphant (2:29:41.960)

because it's one of the,

Lex Fridman (2:29:43.400)

I mentioned my idea for investing directly in open source.

Lex Fridman (2:29:46.760)

So that's a concept called fair OSS.

Lex Fridman (2:29:48.880)

But one of the things we, when we started Quansight,

Travis Oliphant (2:29:51.000)

we knew we would do is we develop products and ideas

Lex Fridman (2:29:53.680)

and new companies might come out.

Lex Fridman (2:29:55.440)

At Anaconda, this was clear, right?

Lex Fridman (2:29:57.680)

Anaconda, we did so much innovation

Travis Oliphant (2:30:00.240)

that like five or six companies could have come out of that.

Lex Fridman (2:30:02.960)

And we just didn't structure it so they could.

Lex Fridman (2:30:05.000)

But in fact, they have, you look at Dask,

Lex Fridman (2:30:07.240)

there's two companies going out of Dask.

Travis Oliphant (2:30:08.880)

You know, Bokeh could be a company.

Lex Fridman (2:30:10.080)

There's like lots of companies that could exist

Travis Oliphant (2:30:11.720)

off the work we did there.

Lex Fridman (2:30:13.120)

And so I thought, oh, here's a recipe for an incubation,

Travis Oliphant (2:30:16.400)

a concept that we could actually spawn new companies

Lex Fridman (2:30:19.480)

and new innovations.

Lex Fridman (2:30:20.800)

And then the idea has always been,

Lex Fridman (2:30:22.800)

well, money they earn should come back

Travis Oliphant (2:30:24.680)

to fund the open source projects.

Lex Fridman (2:30:26.520)

So labs is, you know, I think there should be

Travis Oliphant (2:30:29.240)

a lot of things like Quansight Labs.

Lex Fridman (2:30:30.720)

I think this concept is one that scales.

Travis Oliphant (2:30:32.560)

You could have a lot of open source research labs.

Lex Fridman (2:30:35.080)

Along the way, so in 2018, when the bigger idea came,

Lex Fridman (2:30:37.480)

how to make open source investable, I said,

Lex Fridman (2:30:38.800)

oh, I need to write, I need to create a venture fund.

Lex Fridman (2:30:41.120)

So we created a venture fund called Quansight Initiate

Lex Fridman (2:30:43.840)

at the same time.

Travis Oliphant (2:30:44.680)

It's an angel fund, really.

Lex Fridman (2:30:45.520)

It's, you know, we started to learn that process.

Lex Fridman (2:30:47.840)

How do we actually do this?

Lex Fridman (2:30:48.680)

How do we get LPs?

Lex Fridman (2:30:49.520)

How do we actually go in this direction and build a fund?

Lex Fridman (2:30:52.480)

And I'm like, every venture fund should have

Travis Oliphant (2:30:54.280)

an associated open source research lab,

Lex Fridman (2:30:55.720)

which is no reason.

Travis Oliphant (2:30:56.560)

Like our venture fund, the carried interest,

Lex Fridman (2:30:59.520)

a portion of it goes to the lab.

Travis Oliphant (2:31:01.840)

It directly will fund the lab.

Lex Fridman (2:31:03.280)

That's fascinating, brother.

Lex Fridman (2:31:04.120)

So you use the power of the organic formation of teams

Lex Fridman (2:31:06.800)

in the open source community, and then like naturally,

Travis Oliphant (2:31:10.680)

that leads to a business that can make money.

Lex Fridman (2:31:13.920)

Yeah, correct.

Lex Fridman (2:31:14.760)

And then it always maintains and loops back

Lex Fridman (2:31:16.680)

to the open source.

Travis Oliphant (2:31:17.520)

Loops back to open source, exactly.

Lex Fridman (2:31:18.440)

I mean, to me, it's a natural fit.

Travis Oliphant (2:31:19.640)

There's something, there's absolutely

Lex Fridman (2:31:20.960)

a repeatable pattern there, and it's also beneficial

Travis Oliphant (2:31:23.640)

because, oh, I have, I have natural connections

Lex Fridman (2:31:26.800)

to the open source if I have an open source research lab.

Travis Oliphant (2:31:29.200)

Like, they'll always, they'll be out there

Lex Fridman (2:31:31.160)

talking to people, and so we've had a chance

Travis Oliphant (2:31:34.280)

to talk to a lot of early stage companies.

Lex Fridman (2:31:35.920)

And we, and our fund focuses on the early stage.

Lex Fridman (2:31:37.880)

So Quansight has the services, the lab, the fund, right?

Lex Fridman (2:31:41.880)

In that process, a lot of stuff started to happen.

Travis Oliphant (2:31:44.200)

They're like, oh, you know, we started to do recruiting

Lex Fridman (2:31:46.320)

and support and training, and I was starting

Travis Oliphant (2:31:48.600)

to build a bigger sales team and marketing team

Lex Fridman (2:31:50.960)

and people besides just developers.

Lex Fridman (2:31:52.880)

And one of the challenges with that

Lex Fridman (2:31:54.080)

is you end up with different cultural aspects.

Travis Oliphant (2:31:55.960)

You know, developers, you know, there's a,

Lex Fridman (2:31:58.800)

in any company you go to, you kind of go look,

Lex Fridman (2:32:00.760)

is this a business led company, a developer led company?

Lex Fridman (2:32:03.080)

Do they kind of coexist?

Lex Fridman (2:32:04.280)

Are they, what's the interface between them?

Lex Fridman (2:32:06.120)

There's always a bit of a tension there.

Travis Oliphant (2:32:07.280)

Like we were talking about before.

Lex Fridman (2:32:08.760)

You know, what is the tension there?

Travis Oliphant (2:32:10.200)

With OpenTeams, I thought, wait a minute,

Lex Fridman (2:32:11.360)

we can actually just create,

Travis Oliphant (2:32:13.160)

like this concept of Quansight plus labs,

Lex Fridman (2:32:15.560)

it's, well, it's specific to the Pi data ecosystem.

Travis Oliphant (2:32:18.480)

The concept is general for all open source.

Lex Fridman (2:32:20.800)

So OpenTeams emerged as a, oh,

Travis Oliphant (2:32:22.640)

we can create a business development company

Lex Fridman (2:32:24.400)

for many, many Quansights, like thousands of Quansights.

Lex Fridman (2:32:28.440)

And it can be a marketplace to connect,

Lex Fridman (2:32:30.840)

essentially be the enterprise software company

Travis Oliphant (2:32:33.440)

of the future.

Lex Fridman (2:32:34.440)

If you look at what enterprise software wants

Travis Oliphant (2:32:36.760)

from the customer side, and during this journey,

Lex Fridman (2:32:38.640)

I've had the chance to work and sell to lots of companies,

Travis Oliphant (2:32:42.360)

Exxon and Shell and Davey Morgan Bank of America,

Lex Fridman (2:32:45.240)

like the Fortune 100,

Lex Fridman (2:32:46.680)

and talk to a lot of people in procurement

Lex Fridman (2:32:48.240)

and see what are they buying and why are they buying?

Travis Oliphant (2:32:50.400)

So, you know, I don't know everything,

Lex Fridman (2:32:51.760)

but I've learned a lot about,

Lex Fridman (2:32:52.720)

oh, what are they really looking for?

Lex Fridman (2:32:54.480)

And they're looking for solutions.

Travis Oliphant (2:32:56.400)

They're constantly given products

Lex Fridman (2:32:58.160)

from enterprise software.

Travis Oliphant (2:33:01.160)

Here's open source, leave the enterprise software,

Lex Fridman (2:33:02.560)

now I buy it.

Lex Fridman (2:33:03.400)

And then they have to stitch it together into a solution.

Lex Fridman (2:33:05.880)

Open source is fantastic for gluing

Travis Oliphant (2:33:07.360)

those solutions together.

Lex Fridman (2:33:08.760)

So, whereas they keep getting new platforms

Travis Oliphant (2:33:11.480)

they're trying to buy,

Lex Fridman (2:33:12.360)

but most open source, what most enterprises want

Travis Oliphant (2:33:15.000)

is tools that they can customize

Lex Fridman (2:33:16.800)

that are as inexpensive as they can.

Travis Oliphant (2:33:18.920)

Yeah, and so you always want to maintain

Lex Fridman (2:33:20.400)

the connection to the open source

Travis Oliphant (2:33:21.560)

because that's going to be the tools.

Lex Fridman (2:33:22.400)

Yes, so open teams is about solving

Travis Oliphant (2:33:24.840)

enterprise software problems.

Lex Fridman (2:33:26.720)

Brilliant, brilliant idea, by the way.

Travis Oliphant (2:33:28.120)

With a connect, but we do it honoring the topology.

Lex Fridman (2:33:30.960)

We don't hire all the people.

Travis Oliphant (2:33:32.360)

We are a network connecting the sales energy

Lex Fridman (2:33:35.120)

and the procurement energy,

Lex Fridman (2:33:36.520)

and we work on the business side,

Lex Fridman (2:33:37.960)

get the deals closed,

Lex Fridman (2:33:39.080)

and then have a network of partners

Lex Fridman (2:33:40.560)

like Quonsight and others who we hand the deals to,

Travis Oliphant (2:33:44.080)

to actually do the work.

Lex Fridman (2:33:44.920)

And then we have to maintain,

Travis Oliphant (2:33:46.480)

I feel like we have to maintain

Lex Fridman (2:33:47.320)

some level of quality control

Lex Fridman (2:33:48.760)

so that the client can rely on open teams

Lex Fridman (2:33:50.960)

to ensure the delivery.

Travis Oliphant (2:33:52.080)

It's not just, here's a lead, go figure that out.

Lex Fridman (2:33:54.640)

But no, we're going to make sure you get what you need.

Travis Oliphant (2:33:57.040)

By the way, it's such a skill,

Lex Fridman (2:33:58.840)

and I don't know if I have the patience.

Travis Oliphant (2:34:00.640)

I will have the patience to talk to the business people

Lex Fridman (2:34:04.080)

or more specific, I mean,

Travis Oliphant (2:34:05.600)

there's all kinds of flavors of business people

Lex Fridman (2:34:07.480)

or like marketing people.

Travis Oliphant (2:34:11.960)

There's a challenge.

Lex Fridman (2:34:12.800)

I hear what you're saying

Travis Oliphant (2:34:13.640)

because I've had the same challenge.

Lex Fridman (2:34:14.880)

And it's true.

Travis Oliphant (2:34:15.720)

There's sometimes you think, okay, this is way overwrought.

Lex Fridman (2:34:18.440)

Yeah, but you have to become an adult

Lex Fridman (2:34:20.240)

and you have to, because the companies have needs.

Lex Fridman (2:34:22.320)

They have ways to make money

Lex Fridman (2:34:24.320)

and they also want to learn and grow,

Lex Fridman (2:34:26.480)

and it's your job to kind of educate them on the best way,

Travis Oliphant (2:34:28.960)

like the value of open source, for example.

Lex Fridman (2:34:31.000)

Right, and I'm really grateful for all my experiences

Travis Oliphant (2:34:32.960)

over the past 14 years, understanding that side of it

Lex Fridman (2:34:35.720)

and still learning for sure,

Lex Fridman (2:34:37.160)

but not just understanding from companies,

Lex Fridman (2:34:38.640)

but also dealing with marketing professionals

Lex Fridman (2:34:40.560)

and sales professionals

Lex Fridman (2:34:41.600)

and people that make a career out of that

Lex Fridman (2:34:43.120)

and understanding what they're thinking about

Lex Fridman (2:34:44.360)

and also understanding, well, let's make this better.

Travis Oliphant (2:34:46.840)

We can really make a place.

Lex Fridman (2:34:48.160)

Open teams I see as the transmission layer

Travis Oliphant (2:34:50.480)

between companies and open source communities

Lex Fridman (2:34:53.720)

producing enterprise software solutions.

Travis Oliphant (2:34:55.600)

Eventually we want to,

Lex Fridman (2:34:56.880)

today we're taking on SaaS and MATLAB

Lex Fridman (2:34:59.320)

and tools that we know we can replace for folks.

Lex Fridman (2:35:01.720)

Really, anytime you have a software tool at an organization

Travis Oliphant (2:35:04.560)

where you have to do a lot of customization

Lex Fridman (2:35:06.200)

to make it work for you.

Travis Oliphant (2:35:07.360)

It's not you're just buying this thing off the shelf

Lex Fridman (2:35:09.000)

and it works.

Travis Oliphant (2:35:09.840)

It's like, okay, you buy this system

Lex Fridman (2:35:11.080)

and then you customize it a lot,

Travis Oliphant (2:35:12.840)

usually with expensive consultants

Lex Fridman (2:35:15.280)

to actually make it work for you.

Travis Oliphant (2:35:17.200)

All of those should be replaced by open source foundations

Lex Fridman (2:35:19.760)

with the same customization.

Travis Oliphant (2:35:20.600)

You're doing such important work,

Lex Fridman (2:35:22.360)

such important work in these giant organizations

Travis Oliphant (2:35:25.440)

that do exactly that,

Lex Fridman (2:35:26.520)

taking some proprietary software

Lex Fridman (2:35:28.360)

and hiring a huge team of consultants

Lex Fridman (2:35:30.520)

that customize it and then that whole thing

Travis Oliphant (2:35:32.760)

gets outdated quick.

Lex Fridman (2:35:33.680)

Correct.

Lex Fridman (2:35:34.520)

And so, I mean, that's brilliant.

Lex Fridman (2:35:36.760)

So the one solution to that

Travis Oliphant (2:35:39.360)

is kind of what Tesla's doing a little bit of,

Lex Fridman (2:35:43.240)

which is basically build up a software engineering team.

Travis Oliphant (2:35:46.680)

Like build a team from scratch.

Lex Fridman (2:35:48.320)

Build a team from scratch.

Lex Fridman (2:35:49.160)

And companies are doing it well,

Lex Fridman (2:35:50.000)

that's what they're doing right now.

Travis Oliphant (2:35:50.840)

Yeah, exactly.

Lex Fridman (2:35:51.680)

And that's okay.

Lex Fridman (2:35:52.520)

And you're creating a topology for some of that.

Lex Fridman (2:35:54.360)

You're right.

Travis Oliphant (2:35:55.200)

You just don't have to do it.

Lex Fridman (2:35:56.040)

That's not the only answer, right?

Lex Fridman (2:35:57.040)

And so other companies can access this,

Lex Fridman (2:35:58.880)

be more accessible.

Travis Oliphant (2:35:59.880)

We literally say,

Lex Fridman (2:36:01.120)

open team is the future of enterprise software.

Travis Oliphant (2:36:03.920)

We're still early.

Lex Fridman (2:36:04.760)

Like this idea just percolated over the past year

Travis Oliphant (2:36:07.400)

as we've kind of grown Quansight

Lex Fridman (2:36:08.520)

and realized the extensibility of it.

Travis Oliphant (2:36:10.440)

We just finished in our seed round

Lex Fridman (2:36:13.240)

to help get more sales people

Lex Fridman (2:36:15.160)

and then push the messaging correctly.

Lex Fridman (2:36:17.640)

And there's lots of tools we're building

Travis Oliphant (2:36:19.160)

to make this easier.

Lex Fridman (2:36:20.000)

Like we wanna automate the processes.

Travis Oliphant (2:36:21.720)

We feel like a lot of the power

Lex Fridman (2:36:23.560)

is the efficiency of the sales process.

Travis Oliphant (2:36:25.600)

There's a lot of wasted energy in small teams

Lex Fridman (2:36:29.400)

and the sales energy to get into large companies

Lex Fridman (2:36:31.640)

and make a deal.

Lex Fridman (2:36:32.680)

There's a lot of money spent on that process.

Travis Oliphant (2:36:34.720)

Creating the tools and processes for that sales.

Lex Fridman (2:36:36.560)

So make that super seamless.

Lex Fridman (2:36:38.160)

So a single company can go,

Lex Fridman (2:36:39.680)

oh, I've got my contract with open teams.

Travis Oliphant (2:36:41.400)

We've got a subscription they can get.

Lex Fridman (2:36:43.040)

They can make that procurement seamless.

Lex Fridman (2:36:45.200)

And then the fact they have access

Lex Fridman (2:36:46.720)

to the entire open source ecosystem.

Lex Fridman (2:36:48.840)

And we have a part of our work

Lex Fridman (2:36:51.240)

that's embracing open source ecosystems

Lex Fridman (2:36:53.400)

and making sure we're doing things useful for them

Lex Fridman (2:36:55.080)

or serving them.

Lex Fridman (2:36:56.160)

And then companies making sure

Lex Fridman (2:36:57.560)

they're getting solutions they care about.

Lex Fridman (2:36:59.200)

And then figuring out which targets we have.

Lex Fridman (2:37:02.480)

We're not taking on all of open source,

Travis Oliphant (2:37:04.720)

all of enterprise software yet.

Lex Fridman (2:37:06.040)

But we're step by step.

Travis Oliphant (2:37:07.440)

Well this feels like the future.

Lex Fridman (2:37:08.520)

The idea and the vision is brilliant.

Travis Oliphant (2:37:10.600)

Can I ask you, why do you think Microsoft bought GitHub

Lex Fridman (2:37:14.440)

and what do you think is the future of GitHub?

Travis Oliphant (2:37:16.560)

Great point.

Lex Fridman (2:37:17.400)

I thought it was a brilliant move.

Travis Oliphant (2:37:18.220)

I think they did because Microsoft has always

Lex Fridman (2:37:20.620)

had a developer centric culture.

Travis Oliphant (2:37:22.660)

Like they always have.

Lex Fridman (2:37:23.500)

Like one of the things Microsoft's always done well

Travis Oliphant (2:37:25.160)

is understand that their power is the developers.

Lex Fridman (2:37:27.440)

It's been, Ballmer didn't necessarily make a good meme

Travis Oliphant (2:37:31.600)

about how he approached that.

Lex Fridman (2:37:32.560)

But they're broadening that.

Travis Oliphant (2:37:34.520)

I think that's why.

Lex Fridman (2:37:35.360)

Because they recognize GitHub is where developers are at.

Lex Fridman (2:37:38.080)

Right?

Lex Fridman (2:37:38.920)

And so.

Lex Fridman (2:37:39.740)

But do they have a vision like open teams

Lex Fridman (2:37:41.080)

type of situation, right?

Travis Oliphant (2:37:41.920)

I don't think so yet.

Lex Fridman (2:37:43.600)

Are they just basically throwing money at developers

Lex Fridman (2:37:46.680)

to show their support?

Lex Fridman (2:37:47.960)

I think so.

Travis Oliphant (2:37:48.800)

Without a topology like you put it.

Lex Fridman (2:37:50.840)

Like a way to leverage that.

Travis Oliphant (2:37:53.280)

Like to give developers actual money.

Lex Fridman (2:37:55.480)

Right.

Travis Oliphant (2:37:56.320)

I don't think so.

Lex Fridman (2:37:57.160)

They're still, it's an enterprise software company.

Lex Fridman (2:37:59.440)

And they make a bunch of money.

Lex Fridman (2:38:00.520)

They make a bunch of games.

Travis Oliphant (2:38:01.360)

They're a big company.

Lex Fridman (2:38:02.640)

They sell products.

Travis Oliphant (2:38:03.760)

I think part of it is they know there's opportunity

Lex Fridman (2:38:06.080)

to make money from GitHub.

Lex Fridman (2:38:07.760)

Right?

Lex Fridman (2:38:08.600)

There's definitely a business there.

Travis Oliphant (2:38:09.440)

You know, to sell to developers.

Lex Fridman (2:38:11.340)

Or to sell to people using development.

Travis Oliphant (2:38:13.280)

I think there's part of that.

Lex Fridman (2:38:14.240)

I think part of it is also there's,

Travis Oliphant (2:38:15.880)

they had definitely wanted to recognize

Lex Fridman (2:38:18.080)

that you need to value open source

Travis Oliphant (2:38:20.560)

to get great developers.

Lex Fridman (2:38:21.920)

Which is an important concept that was emerging

Travis Oliphant (2:38:24.000)

over the past 10 years.

Lex Fridman (2:38:25.000)

That, you know, pay at Pi Data.

Travis Oliphant (2:38:28.000)

We were able to convince J.P. Morgan

Lex Fridman (2:38:29.880)

to support Pi Data because of that fact.

Lex Fridman (2:38:31.480)

Right?

Lex Fridman (2:38:32.320)

That was where the money for them putting

Travis Oliphant (2:38:33.440)

a couple hundred thousand into supporting Pi Data

Lex Fridman (2:38:35.160)

for several conferences was they want developers.

Lex Fridman (2:38:37.800)

And they realized that developers want

Lex Fridman (2:38:39.480)

to participate in open source.

Lex Fridman (2:38:40.720)

So enterprise software folks don't always understand

Lex Fridman (2:38:43.200)

how their software gets used.

Travis Oliphant (2:38:44.600)

Having spent a lot of time on the floors

Lex Fridman (2:38:46.560)

at J.P. Morgan, at InShell, at ExxonMobil,

Travis Oliphant (2:38:49.600)

you see, oh, these companies have large development teams.

Lex Fridman (2:38:52.880)

And then they're kind of dealing with

Travis Oliphant (2:38:55.280)

what's being delivered to them.

Lex Fridman (2:38:56.720)

So I really feel kind of a privilege

Travis Oliphant (2:38:58.360)

that I had a chance to learn some of these people

Lex Fridman (2:39:00.480)

and see what they're doing.

Lex Fridman (2:39:01.800)

And even work alongside them, you know,

Lex Fridman (2:39:04.000)

as a consultant, using open source and trying to figure,

Lex Fridman (2:39:07.640)

how do we make this work inside of our large organization?

Lex Fridman (2:39:09.960)

Some of it is actually, for a large organization,

Travis Oliphant (2:39:13.000)

some of it is messaging to the world

Lex Fridman (2:39:14.800)

that you care about developers

Lex Fridman (2:39:16.280)

and you're the cool, you care.

Lex Fridman (2:39:18.840)

Like, for example, like if Ford,

Lex Fridman (2:39:21.040)

cause I talked to them, like car companies, right?

Lex Fridman (2:39:23.880)

They want to attract, you know,

Travis Oliphant (2:39:26.680)

you want to take on Tesla and autopilot.

Lex Fridman (2:39:28.760)

You want to take on, right?

Lex Fridman (2:39:29.960)

And so what do you do there?

Lex Fridman (2:39:31.720)

You show that you're cool.

Travis Oliphant (2:39:32.960)

Like you try to show off that you care about developers

Lex Fridman (2:39:36.480)

and they have a lot of trouble doing that.

Lex Fridman (2:39:39.040)

And like one way, I think like Ford should have bought GitHub.

Lex Fridman (2:39:42.720)

They just to show off, like these old school companies

Lex Fridman (2:39:46.880)

and it's in a lot of different industries.

Lex Fridman (2:39:49.960)

There's probably different ways.

Travis Oliphant (2:39:51.080)

It's probably an art show that you care to developers.

Lex Fridman (2:39:54.080)

And the developers, it's exactly what you, like,

Travis Oliphant (2:39:57.920)

for example, just spit balling here,

Lex Fridman (2:40:00.520)

but like Ford or somebody like that

Travis Oliphant (2:40:02.520)

could give a hundred million dollars

Lex Fridman (2:40:05.960)

to the development of NumPy.

Lex Fridman (2:40:07.880)

And like literally look at like the top most popular projects

Lex Fridman (2:40:13.200)

in Python and just say, we're just going to give money.

Travis Oliphant (2:40:17.080)

Like that's going to immediately make you cool.

Lex Fridman (2:40:20.240)

They could actually, yeah.

Lex Fridman (2:40:21.600)

And in fact, they set up NumFocus to make it easy.

Lex Fridman (2:40:24.400)

But the challenge was,

Travis Oliphant (2:40:26.080)

is also you have to have some business development.

Lex Fridman (2:40:28.480)

Like it's a bit of a seeding problem, right?

Lex Fridman (2:40:31.280)

And you look at how,

Lex Fridman (2:40:32.120)

I've talked to the folks at Linux Foundation,

Travis Oliphant (2:40:33.400)

know how they're doing it.

Lex Fridman (2:40:34.360)

I know how, and starting NumFocus,

Travis Oliphant (2:40:36.600)

because we had two babies in 2012.

Lex Fridman (2:40:39.400)

One was Anaconda, one was NumFocus, right?

Lex Fridman (2:40:41.120)

And they were both important efforts.

Lex Fridman (2:40:42.760)

They had distinct journeys

Lex Fridman (2:40:44.000)

and super grateful that both existed

Lex Fridman (2:40:46.200)

and still grateful both exist.

Lex Fridman (2:40:48.720)

But there's different energies in getting donations

Lex Fridman (2:40:51.840)

as there is getting, this is important to my business.

Travis Oliphant (2:40:55.320)

Like I'm selling you something that this is a,

Lex Fridman (2:40:58.680)

I'm going to make money this way.

Travis Oliphant (2:41:00.280)

Like if you can tie it,

Lex Fridman (2:41:01.120)

if you can tie the message to an ROI for the company,

Travis Oliphant (2:41:04.040)

it becomes a brainer.

Lex Fridman (2:41:04.880)

That's more effective.

Lex Fridman (2:41:05.720)

It's much more effective, right?

Lex Fridman (2:41:06.920)

So, and there are rational arguments to make.

Travis Oliphant (2:41:09.520)

I've tried to have conversations with marketing,

Lex Fridman (2:41:11.120)

especially marketing departments.

Travis Oliphant (2:41:12.240)

Like very early on, it was clear to me that,

Lex Fridman (2:41:14.840)

oh, you could just take a fraction of your marketing budget

Lex Fridman (2:41:18.160)

and just spend it on open source development.

Lex Fridman (2:41:20.240)

And you get better results from your marketing.

Travis Oliphant (2:41:23.760)

Like, because.

Lex Fridman (2:41:24.600)

How did those, can I, sorry,

Travis Oliphant (2:41:26.000)

I'm going to try not to go and rants here.

Lex Fridman (2:41:27.920)

What have you learned from the interaction

Travis Oliphant (2:41:29.800)

with the marketing folks on that kind of,

Lex Fridman (2:41:31.440)

because you gave a great example

Travis Oliphant (2:41:34.160)

of something that will obviously be much better investment

Lex Fridman (2:41:37.240)

in terms of marketing is supporting open source projects.

Travis Oliphant (2:41:40.360)

The challenge is not dissimilar

Lex Fridman (2:41:41.840)

from the challenge you have in academia

Lex Fridman (2:41:44.480)

or the different colleges, right?

Lex Fridman (2:41:46.520)

Knowledge gets very specific and very channeled, right?

Lex Fridman (2:41:50.000)

And so people get,

Lex Fridman (2:41:51.160)

they get a lot of learning in the thing they know about.

Lex Fridman (2:41:53.920)

And it's hard then to bridge that

Lex Fridman (2:41:56.200)

and to get them to think differently enough

Travis Oliphant (2:41:58.160)

to have a sense that you might have something to offer

Lex Fridman (2:42:02.160)

because it's different.

Lex Fridman (2:42:03.000)

It's like, well, how do I implement that?

Lex Fridman (2:42:04.280)

How do I, what do I do with that?

Lex Fridman (2:42:05.840)

Like, do I, which budget do I take from?

Lex Fridman (2:42:07.840)

Do I slow down my spend on Google ads

Lex Fridman (2:42:10.320)

or my spend on Facebook ads?

Lex Fridman (2:42:11.600)

Or do I not hire a content creator and say like,

Travis Oliphant (2:42:14.640)

there's an operational aspect to that,

Lex Fridman (2:42:16.160)

that you have to be the CMO, right?

Travis Oliphant (2:42:19.080)

Or the CEO, you have to get the right level.

Lex Fridman (2:42:21.000)

So you'll have to hire at a high position level

Travis Oliphant (2:42:24.360)

where they care about this and this.

Lex Fridman (2:42:25.720)

Right, or they won't know how, right?

Lex Fridman (2:42:27.640)

And because you can also do it very clumsily, right?

Lex Fridman (2:42:30.440)

And I've seen it, cause you can,

Travis Oliphant (2:42:32.040)

you absolutely have to honor and recognize

Lex Fridman (2:42:33.760)

the people you're going to and the fact

Travis Oliphant (2:42:36.640)

that if you just throw money at them,

Lex Fridman (2:42:37.800)

it could actually create more problems.

Travis Oliphant (2:42:39.240)

Can I just say, this is not you saying, can I just,

Lex Fridman (2:42:41.320)

cause I just need, I need to say this.

Travis Oliphant (2:42:44.360)

I've been very surprised how often marketing people

Lex Fridman (2:42:49.880)

are terrible at marketing.

Travis Oliphant (2:42:51.760)

I feel like the best marketing is doing something novel

Lex Fridman (2:42:55.600)

and unique that anticipates the future.

Travis Oliphant (2:42:58.240)

It feels like so much of the marketing practice

Lex Fridman (2:43:01.520)

is like what they took in school,

Travis Oliphant (2:43:04.320)

or maybe they're studying for what was the best thing

Lex Fridman (2:43:06.680)

that was done in the past decade,

Lex Fridman (2:43:08.440)

and they're just repeating that over and over,

Lex Fridman (2:43:10.800)

as opposed to innovating, like taking the risk.

Travis Oliphant (2:43:13.760)

To me, marketing.

Lex Fridman (2:43:14.600)

That's a great point.

Travis Oliphant (2:43:15.440)

Is taking the big risk.

Lex Fridman (2:43:17.080)

That's a great point.

Lex Fridman (2:43:17.920)

And being the first one to risk.

Lex Fridman (2:43:18.800)

Yeah, there's an aspect of data observation

Lex Fridman (2:43:21.200)

from that risk, right?

Lex Fridman (2:43:22.160)

That's, I think, shared what they're doing already.

Lex Fridman (2:43:25.120)

But it absolutely, it's about, I think it's content.

Lex Fridman (2:43:27.680)

Like there's this whole world on content marketing

Travis Oliphant (2:43:30.200)

that you could almost say, well, yeah, it can get over,

Lex Fridman (2:43:33.560)

you can get inundated with stuff

Travis Oliphant (2:43:35.080)

that's not relevant to you.

Lex Fridman (2:43:36.400)

Whereas what you're saying would be highly relevant

Lex Fridman (2:43:39.160)

and highly useful and highly beneficial.

Lex Fridman (2:43:41.560)

Yeah, but it's risk.

Travis Oliphant (2:43:42.960)

I mean, that's why I sort of,

Lex Fridman (2:43:44.600)

there's a lot of innovative ways of doing that.

Travis Oliphant (2:43:46.240)

Tesla's an example of people

Lex Fridman (2:43:48.000)

that basically don't do marketing.

Travis Oliphant (2:43:49.960)

They do marketing in a very, like,

Lex Fridman (2:43:52.800)

let's say Elon hired a person who's just good at Twitter

Travis Oliphant (2:43:55.720)

for running Tesla's Twitter account.

Lex Fridman (2:43:57.520)

No, right, right.

Travis Oliphant (2:43:59.120)

I mean, that's exactly what you wanna be doing.

Lex Fridman (2:44:00.840)

You want it to be constantly innovating in the.

Travis Oliphant (2:44:03.120)

Right, there's an aspect of telling.

Lex Fridman (2:44:04.280)

I mean, I've definitely seen people doing great work

Travis Oliphant (2:44:06.920)

where you're not talking about it.

Lex Fridman (2:44:08.400)

Like, I would say that's actually a problem

Travis Oliphant (2:44:09.560)

I have right now with Quonset Labs.

Lex Fridman (2:44:11.360)

Quonset Labs has been doing amazing work,

Travis Oliphant (2:44:12.720)

really excited about it,

Lex Fridman (2:44:13.560)

but we have not been talking about it enough.

Travis Oliphant (2:44:15.480)

We haven't been.

Lex Fridman (2:44:16.320)

And there's different ways to talk about it.

Travis Oliphant (2:44:17.880)

There's different ways to,

Lex Fridman (2:44:18.720)

there's different channels to which to communicate.

Travis Oliphant (2:44:20.800)

There's also, like, I'll just throw some shade

Lex Fridman (2:44:25.600)

at companies I love.

Lex Fridman (2:44:27.880)

So for example, iRobot,

Lex Fridman (2:44:29.160)

I just had a conversation with them.

Travis Oliphant (2:44:30.800)

They make Roombas.

Lex Fridman (2:44:31.840)

Sure.

Lex Fridman (2:44:32.680)

And I think I love, they're incredible robots,

Lex Fridman (2:44:35.440)

but like every time they do like advertisement,

Travis Oliphant (2:44:38.960)

not advertisement, but like marketing type stuff,

Lex Fridman (2:44:41.880)

it just looks so corporate.

Lex Fridman (2:44:44.080)

And to me, the incredible,

Lex Fridman (2:44:47.640)

maybe wrong in the case of iRobot, I don't know.

Lex Fridman (2:44:50.280)

But to me, when you're talking about engineering systems,

Lex Fridman (2:44:54.000)

it's really nice to show off the magic of the engineering

Lex Fridman (2:44:57.000)

and the software and all the geniuses behind this product

Lex Fridman (2:45:02.000)

and the tinkering and like the raw authenticity

Travis Oliphant (2:45:05.080)

of what it takes to build that system

Lex Fridman (2:45:06.800)

versus the marketing people who want to have like

Travis Oliphant (2:45:09.960)

pretty people, like standing there all pretty

Lex Fridman (2:45:12.120)

with the robots, like moving perfectly.

Lex Fridman (2:45:14.600)

So to me, there's some aspect,

Lex Fridman (2:45:16.520)

it's like speaking to the hackers,

Travis Oliphant (2:45:18.040)

you have to throw some bones,

Lex Fridman (2:45:21.040)

some care towards the engineers, the developers,

Travis Oliphant (2:45:25.560)

because there's some aspect, one, for the hiring,

Lex Fridman (2:45:28.720)

but two, there's an authenticity to that,

Travis Oliphant (2:45:31.000)

authenticity to that kind of communication

Lex Fridman (2:45:33.080)

that's really inspiring to the end user as well.

Travis Oliphant (2:45:36.080)

Like if they know that brilliant people,

Lex Fridman (2:45:38.440)

the best in the world are working at your company,

Travis Oliphant (2:45:40.680)

they start to believe that that product

Lex Fridman (2:45:42.640)

that you're creating is really good.

Travis Oliphant (2:45:43.960)

It's interesting, because your initial reaction would be,

Lex Fridman (2:45:45.640)

wait, there's different users here.

Lex Fridman (2:45:46.760)

Why would you do that to, you know,

Lex Fridman (2:45:48.400)

my wife bought a Roomba, and she loves developers,

Travis Oliphant (2:45:52.120)

she loves me, but she doesn't care about that culture.

Lex Fridman (2:45:56.560)

So essentially what you said is actually the authenticity,

Travis Oliphant (2:45:59.600)

because everyone has a friend, everyone knows people,

Lex Fridman (2:46:01.160)

there's word of mouth, I mean, if you.

Travis Oliphant (2:46:02.680)

Word of mouth is so, so proper.

Lex Fridman (2:46:04.160)

Yeah, exactly, that's interesting.

Travis Oliphant (2:46:05.640)

Because I think it's the lack of that realization,

Lex Fridman (2:46:07.560)

there's this halo effect that influences

Travis Oliphant (2:46:09.840)

your general marketing, interesting.

Lex Fridman (2:46:11.720)

For some stupid reason, I do have a platform,

Lex Fridman (2:46:14.640)

and it seems that the reason I have a platform,

Lex Fridman (2:46:16.920)

many others like me, millions of others,

Travis Oliphant (2:46:19.480)

is like the authenticity,

Lex Fridman (2:46:21.160)

and like we get excited naturally about stuff.

Lex Fridman (2:46:23.960)

And like, I don't want to get excited

Lex Fridman (2:46:25.760)

about that iRobot video,

Travis Oliphant (2:46:27.800)

because it's boring, it's marketing, it's corporate,

Lex Fridman (2:46:30.760)

as opposed to, I wanted to do some fun,

Travis Oliphant (2:46:33.600)

this is me, like a shout out to iRobot,

Lex Fridman (2:46:36.240)

is they're not letting me get into the robot.

Travis Oliphant (2:46:39.360)

Yeah, well there's an aspect of,

Lex Fridman (2:46:40.920)

that could be benefiting from a culture of modularity,

Travis Oliphant (2:46:44.840)

like add ons, and that could actually dramatically help.

Lex Fridman (2:46:47.840)

You've seen that over history,

Travis Oliphant (2:46:49.300)

I mean, Apple is an example of a company like that,

Lex Fridman (2:46:51.160)

or the, like, I can see what your point is,

Travis Oliphant (2:46:54.400)

is that you have something that needs to be,

Lex Fridman (2:46:56.920)

it needs to be adopted broadly,

Travis Oliphant (2:46:58.240)

the concept needs to be adopted broadly.

Lex Fridman (2:47:00.040)

And if you want to go beyond this one device,

Travis Oliphant (2:47:01.640)

you need to engage this community.

Lex Fridman (2:47:04.220)

Yeah, and connecting to the open source that you said.

Travis Oliphant (2:47:07.560)

I gotta ask you,

Lex Fridman (2:47:09.960)

you're a programmer,

Travis Oliphant (2:47:11.800)

one of the most impactful programmers ever.

Lex Fridman (2:47:14.840)

You've led many programmers, you lead many programmers.

Lex Fridman (2:47:18.560)

What are some, from a programmer perspective,

Lex Fridman (2:47:21.180)

what makes a good programmer?

Lex Fridman (2:47:23.360)

What makes a productive programmer?

Lex Fridman (2:47:25.000)

Is there a device you can give

Lex Fridman (2:47:27.140)

to be a great programmer in this world?

Lex Fridman (2:47:28.480)

That's a great, great question.

Lex Fridman (2:47:30.280)

And there are times in my life

Lex Fridman (2:47:31.640)

I'd probably answer this even better

Travis Oliphant (2:47:32.920)

than I hope maybe give an answer today.

Lex Fridman (2:47:35.040)

Because I thought about this numerous times,

Travis Oliphant (2:47:36.700)

like right now I've spent on so much time

Lex Fridman (2:47:38.280)

recently hiring salespeople that,

Travis Oliphant (2:47:41.000)

That your mind is a little bit on something else.

Lex Fridman (2:47:43.440)

On something else.

Lex Fridman (2:47:44.280)

But I reflected on the past,

Lex Fridman (2:47:46.080)

and also, you know, I have some really,

Travis Oliphant (2:47:48.160)

the only way I can do this,

Lex Fridman (2:47:49.000)

is I have some really great programmers that I work with,

Travis Oliphant (2:47:51.440)

who lead the teams that they lead.

Lex Fridman (2:47:53.240)

And my goal is to inspire them and hopefully help them,

Travis Oliphant (2:47:56.600)

encourage them, and be,

Lex Fridman (2:47:57.800)

help them encourage with their teams.

Travis Oliphant (2:47:59.620)

I would say there's a number of things, couple things.

Lex Fridman (2:48:01.200)

One is curiosity.

Travis Oliphant (2:48:03.860)

Like you, I think a programmer without curiosity

Lex Fridman (2:48:07.700)

is mundane.

Travis Oliphant (2:48:09.600)

Like you'll lose interest, you won't do your best work.

Lex Fridman (2:48:12.240)

So it's sort of, it's an affect.

Travis Oliphant (2:48:13.640)

It's sort of, are you,

Lex Fridman (2:48:14.480)

you have some curiosity about things.

Travis Oliphant (2:48:16.800)

I think two, don't try to do everything at once.

Lex Fridman (2:48:19.600)

Recognize that you're, you know, we're limited as humans.

Travis Oliphant (2:48:21.960)

You're limited as a human.

Lex Fridman (2:48:23.200)

And each one of us are limited in different ways.

Travis Oliphant (2:48:24.920)

You know, we all have our different strengths and skills.

Lex Fridman (2:48:26.600)

So it's adapting the art of programming to your skills.

Travis Oliphant (2:48:29.880)

One of the things that always works,

Lex Fridman (2:48:31.240)

is to limit what you're trying to solve.

Travis Oliphant (2:48:33.580)

Right, so, if you're part of a team,

Lex Fridman (2:48:36.640)

usually maybe somebody else has put the architecture together

Lex Fridman (2:48:38.920)

and they've gotten given a portion for you if you're young.

Lex Fridman (2:48:41.720)

If you're not part of a team,

Travis Oliphant (2:48:43.440)

it's sort of breaking down the problem into smaller parts,

Lex Fridman (2:48:46.640)

is essential for you to make progress.

Travis Oliphant (2:48:48.620)

It's very easy to take on a big project

Lex Fridman (2:48:50.720)

and try to do it all at once, and you get lost.

Lex Fridman (2:48:52.800)

And then you do it badly.

Lex Fridman (2:48:53.680)

And so thinking about, you know,

Travis Oliphant (2:48:57.700)

very concretely what you're doing,

Lex Fridman (2:48:59.400)

defining the inputs and outputs,

Travis Oliphant (2:49:01.440)

defining what you want to get done.

Lex Fridman (2:49:03.960)

Even just talking about that and like writing down

Lex Fridman (2:49:07.280)

before you write code, just what are you trying to accomplish?

Lex Fridman (2:49:09.440)

I mean, very specific about it, really, really helps.

Lex Fridman (2:49:12.800)

I think using other people's work, right?

Lex Fridman (2:49:17.000)

Don't be afraid that somehow you're,

Travis Oliphant (2:49:20.000)

like you should do it all.

Lex Fridman (2:49:21.280)

Like, nobody does.

Travis Oliphant (2:49:23.240)

Stand on the shoulders of giants.

Lex Fridman (2:49:25.240)

And copy and paste from Stack Overflow.

Travis Oliphant (2:49:26.720)

Copy and paste from Stack Overflow.

Lex Fridman (2:49:28.200)

But don't just copy and paste,

Travis Oliphant (2:49:30.040)

this is particularly relevant in the era of Codex

Lex Fridman (2:49:31.760)

and the auto generated code, which is essentially,

Travis Oliphant (2:49:34.960)

I see as an indexing of Stack Overflow.

Lex Fridman (2:49:36.760)

Right, exactly.

Travis Oliphant (2:49:37.600)

Secondly, it's like.

Lex Fridman (2:49:38.440)

It's a search engine.

Travis Oliphant (2:49:39.280)

It's a search engine over Stack Overflow, basically.

Lex Fridman (2:49:41.280)

So it's not, I mean, we've had this for a while.

Lex Fridman (2:49:43.480)

But really, you want to cut and paste, but not blindly.

Lex Fridman (2:49:47.300)

Like, absolutely I've cut and paste to understand,

Lex Fridman (2:49:51.000)

but then you understand.

Lex Fridman (2:49:52.320)

Oh, this is what this means.

Travis Oliphant (2:49:53.640)

Oh, this is what it's doing.

Lex Fridman (2:49:54.920)

And understand as much as you can.

Lex Fridman (2:49:56.680)

So it's critical, that's where the curiosity comes in.

Lex Fridman (2:49:59.080)

If you're just blindly cutting and pasting,

Travis Oliphant (2:50:01.000)

you're not gonna understand.

Lex Fridman (2:50:02.240)

So understand, and then be sensitive to hype cycles.

Travis Oliphant (2:50:08.600)

Right, every few often there's always a,

Lex Fridman (2:50:10.920)

oh, test driven development is the answer.

Travis Oliphant (2:50:12.520)

Oh, object oriented is the answer.

Lex Fridman (2:50:13.800)

Oh, there's always an answer.

Travis Oliphant (2:50:16.520)

Agile is the answer.

Lex Fridman (2:50:18.400)

Be cautious of jumping onto a hype cycle.

Travis Oliphant (2:50:20.840)

Like, likely there's signal.

Lex Fridman (2:50:22.520)

Like, there's a thing there

Travis Oliphant (2:50:23.440)

that's actually valuable, you can learn from.

Lex Fridman (2:50:25.320)

But it's almost certainly not the answer

Travis Oliphant (2:50:27.720)

to everything you need.

Lex Fridman (2:50:28.960)

What lessons do you draw

Lex Fridman (2:50:30.160)

from you having created NumPy and SciPy?

Lex Fridman (2:50:34.100)

Like, in service of sort of answering the question

Travis Oliphant (2:50:37.200)

of what it takes to be a great programmer

Lex Fridman (2:50:38.840)

and giving advice to people.

Lex Fridman (2:50:40.520)

How can you be the next person to create a SciPy?

Lex Fridman (2:50:42.960)

Yeah, so one is listen.

Lex Fridman (2:50:45.640)

To?

Lex Fridman (2:50:46.480)

Listen.

Lex Fridman (2:50:47.300)

To who?

Lex Fridman (2:50:48.140)

To people that have a problem, right?

Lex Fridman (2:50:51.440)

Which is everybody, right?

Lex Fridman (2:50:52.520)

But listen, and listen to many.

Lex Fridman (2:50:54.960)

And then try to, and then do.

Lex Fridman (2:50:57.460)

Like, you're gonna have to do an experiment, you know?

Travis Oliphant (2:50:59.760)

Do, fall down, don't be afraid to fall down.

Lex Fridman (2:51:01.940)

Don't be afraid, the first thing you do

Lex Fridman (2:51:04.240)

is probably gonna suck, and that's okay, right?

Lex Fridman (2:51:07.600)

It's honestly, I think iteration is the key to innovation.

Lex Fridman (2:51:11.240)

And it's almost that psychological hesitation we have

Lex Fridman (2:51:16.240)

to just iterate.

Travis Oliphant (2:51:18.520)

Like, yeah, we know it's not great,

Lex Fridman (2:51:20.560)

but next time it'll be better.

Travis Oliphant (2:51:22.000)

I mean, just keep learning and keep improving.

Lex Fridman (2:51:25.560)

So it's an attitude.

Lex Fridman (2:51:27.700)

And then it doesn't take intense concentration, right?

Lex Fridman (2:51:32.160)

Good things don't happen just,

Lex Fridman (2:51:34.560)

it's not quite like TikTok or like Facebook, you know?

Lex Fridman (2:51:38.200)

You can't scroll your way to good programming, right?

Travis Oliphant (2:51:40.520)

There are sincere hours of deep,

Lex Fridman (2:51:44.720)

don't be afraid of the deep problem.

Travis Oliphant (2:51:46.040)

Like, often people will run away from something

Lex Fridman (2:51:47.680)

because, oh, I can't solve this.

Lex Fridman (2:51:49.000)

And you might be right, but give it an hour.

Lex Fridman (2:51:51.360)

Give it a couple of hours and see.

Lex Fridman (2:51:53.360)

And just five minutes, not gonna give you that.

Lex Fridman (2:51:56.560)

Was it lonely when you were building SciPy and NumPy?

Travis Oliphant (2:52:00.520)

Hugely, yeah, absolutely lonely,

Lex Fridman (2:52:02.520)

in the sense of you had to have an inner drive,

Lex Fridman (2:52:05.760)

and that inner drive for me always comes from,

Lex Fridman (2:52:08.000)

I have to see that this is right in some angle.

Travis Oliphant (2:52:11.640)

I have to believe it, that this is the right approach,

Lex Fridman (2:52:13.360)

the right thing to do.

Travis Oliphant (2:52:14.720)

With SciPy, it was like, oh yeah,

Lex Fridman (2:52:16.400)

the world needs libraries and Python.

Travis Oliphant (2:52:19.080)

Clearly Python's popular enough

Lex Fridman (2:52:20.720)

with enough influential people to start,

Lex Fridman (2:52:22.960)

and it needs more libraries.

Lex Fridman (2:52:24.640)

So that is a good in and of itself.

Lex Fridman (2:52:26.600)

So I'm gonna go do that good.

Lex Fridman (2:52:28.360)

So find a good, find a thing that you know is good

Lex Fridman (2:52:30.360)

and just work on it.

Lex Fridman (2:52:33.040)

So that has to happen, and it is.

Lex Fridman (2:52:34.720)

And you kind of have to have enough realization

Lex Fridman (2:52:37.000)

of your mission to be okay with the naysayer

Travis Oliphant (2:52:40.280)

or the fact that not everybody joins you at front.

Lex Fridman (2:52:42.200)

In fact, one thing I've talked to people a lot,

Travis Oliphant (2:52:43.480)

I've seen a lot of projects come, and some fail.

Lex Fridman (2:52:45.480)

Not everything I've done has actually worked perfectly.

Travis Oliphant (2:52:47.600)

I've tried a bunch of stuff that, okay,

Lex Fridman (2:52:49.160)

that didn't really work, or this isn't working, and why.

Lex Fridman (2:52:51.920)

But you see the patterns, and one of the key things is

Lex Fridman (2:52:55.800)

you can't even know for six months.

Travis Oliphant (2:52:59.040)

I say 18 months right now.

Lex Fridman (2:53:00.200)

If you're starting a new project,

Travis Oliphant (2:53:01.800)

you gotta give it a good 18 month run

Lex Fridman (2:53:03.200)

before you even know if the feedback's there.

Travis Oliphant (2:53:05.920)

You're not gonna know in six months.

Lex Fridman (2:53:07.880)

You might have the perfect thing,

Lex Fridman (2:53:08.720)

but six months from now, it's still kind of still emerging.

Lex Fridman (2:53:11.480)

So give it time, because you're dealing with humans,

Lex Fridman (2:53:13.360)

and humans have an inertial energy

Lex Fridman (2:53:15.960)

that just doesn't change that quickly, so.

Travis Oliphant (2:53:18.680)

Let me ask a silly question, but like you said,

Lex Fridman (2:53:23.560)

you're focused on the sales side of things currently,

Lex Fridman (2:53:26.120)

but back when you were actively programming,

Lex Fridman (2:53:28.960)

maybe in the 90s, you talked about IDEs.

Lex Fridman (2:53:31.680)

What's a setup that you have that brings you joy?

Lex Fridman (2:53:36.200)

Keyboard, number of screens, Linux.

Travis Oliphant (2:53:39.640)

I do still like to program some.

Lex Fridman (2:53:40.920)

It's not as much as I used to.

Travis Oliphant (2:53:42.160)

I have two projects I'm super interested in,

Lex Fridman (2:53:44.560)

trying to find funding for them,

Travis Oliphant (2:53:45.640)

trying to figure out teams for them,

Lex Fridman (2:53:47.200)

but I could talk about those.

Lex Fridman (2:53:49.040)

But what I, yeah, I'm an Emacs guy.

Lex Fridman (2:53:51.960)

Great, thank the superior editor, everybody.

Travis Oliphant (2:53:56.080)

I've got, I don't often delete tweets,

Lex Fridman (2:53:59.000)

but one of the tweets I deleted

Travis Oliphant (2:54:00.600)

when I said Emacs was better than Vim,

Lex Fridman (2:54:02.840)

and then the hate I got from it.

Travis Oliphant (2:54:04.520)

It is.

Lex Fridman (2:54:05.360)

I was like, I'm walking away from this.

Travis Oliphant (2:54:07.640)

I do too, I don't push it.

Lex Fridman (2:54:09.160)

I mean, I'm not.

Travis Oliphant (2:54:10.000)

I'm just joking, of course.

Lex Fridman (2:54:11.080)

Yeah, exactly, it's kind of like,

Lex Fridman (2:54:12.160)

but people do take the editor seriously, right?

Lex Fridman (2:54:14.520)

I did it as a joke.

Travis Oliphant (2:54:15.360)

That's your life.

Lex Fridman (2:54:16.200)

It is, but there's something beautiful to me about Emacs,

Lex Fridman (2:54:20.760)

but for people that love Vim,

Lex Fridman (2:54:22.360)

there's something beautiful to them about that.

Travis Oliphant (2:54:23.200)

There is.

Lex Fridman (2:54:24.040)

I mean, I do use Vim for quick editing.

Travis Oliphant (2:54:26.280)

Like Command Line, if I said quick editing,

Lex Fridman (2:54:27.880)

I will still sometimes use it, but not much.

Travis Oliphant (2:54:30.280)

Like it's simple, corrective signal editor character.

Lex Fridman (2:54:32.760)

So when you were developing SciPy, you were using Emacs?

Travis Oliphant (2:54:34.920)

Emacs, yeah.

Lex Fridman (2:54:35.880)

SciPy and NumPy are all written on Emacs on a Linux box.

Lex Fridman (2:54:39.040)

And CVS and then SVN, version control.

Lex Fridman (2:54:43.160)

Git came later.

Travis Oliphant (2:54:44.040)

Like Git has, I love distributed branch stuff.

Lex Fridman (2:54:48.080)

I think Git is pretty complicated, but I love the concept.

Lex Fridman (2:54:51.640)

And also, of course, GitHub and then GitLab

Lex Fridman (2:54:55.240)

make Git definitely consumable, but that came later.

Lex Fridman (2:54:59.440)

Did you ever touch Lisp at all?

Lex Fridman (2:55:00.880)

Like what were your emotional feelings

Lex Fridman (2:55:03.400)

about all the parentheses?

Lex Fridman (2:55:04.240)

Yeah, so great question.

Lex Fridman (2:55:05.440)

So I find myself appreciating Lisp today

Lex Fridman (2:55:08.240)

much more than I did early.

Travis Oliphant (2:55:09.680)

Because when I came to programming, I knew programming,

Lex Fridman (2:55:11.680)

but I was a domain expert, right?

Lex Fridman (2:55:13.000)

And to me, the parentheses were in the way.

Lex Fridman (2:55:15.720)

It's like, wow, there's just all this,

Travis Oliphant (2:55:17.800)

like it just gets in the way of my thinking

Lex Fridman (2:55:19.320)

about what I'm doing.

Lex Fridman (2:55:20.160)

So why would I have all these, right?

Lex Fridman (2:55:22.440)

That was my initial reaction to it.

Lex Fridman (2:55:24.760)

And now as I appreciate kind of the structure

Lex Fridman (2:55:27.320)

that kind of naturally maps to a logical thinking

Lex Fridman (2:55:30.280)

about a program, I can appreciate them, right?

Lex Fridman (2:55:33.000)

And why it's actually, you could create editors

Travis Oliphant (2:55:35.680)

that make it not so problematic, right, honestly.

Lex Fridman (2:55:40.720)

So I actually have a much more appreciation of Lisp

Lex Fridman (2:55:43.000)

and things like Clojure and there's HyVee,

Lex Fridman (2:55:44.720)

which is a Python Lisp that compiles the Python bytecode.

Travis Oliphant (2:55:48.560)

I think it's challenging.

Lex Fridman (2:55:50.280)

Like typically these languages are,

Travis Oliphant (2:55:53.160)

I even saw the whole data science programming system

Lex Fridman (2:55:55.280)

in Lisp that somebody created, which is cool.

Lex Fridman (2:55:58.360)

But again, I think it's the lack of recognition

Lex Fridman (2:56:00.840)

of the fact that there exists

Lex Fridman (2:56:02.020)

what I call occasional programmers.

Lex Fridman (2:56:04.080)

People that are never gonna be programmers for a living.

Travis Oliphant (2:56:05.840)

They don't want to have all this cuteness in their head.

Lex Fridman (2:56:08.440)

They want just, it's why basic, you know,

Travis Oliphant (2:56:11.880)

Microsoft had the right idea with basic

Lex Fridman (2:56:14.480)

in terms of having that be the language of visual basic,

Travis Oliphant (2:56:17.660)

the language of Excel and SQL Server.

Lex Fridman (2:56:21.280)

They should have converted that to Python 10 years ago.

Travis Oliphant (2:56:23.520)

Like the world would be a better place if they had, but.

Lex Fridman (2:56:27.200)

There's also, there's a beauty and a magic

Travis Oliphant (2:56:29.660)

to the history behind a language in Lisp.

Lex Fridman (2:56:31.640)

You know, some of the most interesting people

Travis Oliphant (2:56:34.020)

in the history of computer science

Lex Fridman (2:56:35.880)

and artificial intelligence have used Lisp.

Lex Fridman (2:56:37.920)

So you feel.

Lex Fridman (2:56:40.000)

Well, especially that language,

Travis Oliphant (2:56:41.200)

when you have a language, you can think in it.

Lex Fridman (2:56:43.440)

And it helps you think better.

Lex Fridman (2:56:44.280)

And it attracts a certain kinds of people

Lex Fridman (2:56:45.640)

that think in a certain kind of way.

Lex Fridman (2:56:46.920)

And then that's there.

Lex Fridman (2:56:48.560)

Okay, so what about like small laptop with a tiny keyboard,

Lex Fridman (2:56:52.140)

or is there like three screens?

Lex Fridman (2:56:55.000)

You know, good question.

Travis Oliphant (2:56:55.840)

I've never gotten into the big, many screens to be honest.

Lex Fridman (2:56:58.080)

I mean, and maybe it's because in my head,

Travis Oliphant (2:57:00.720)

I kind of just, I just swap between windows.

Lex Fridman (2:57:03.480)

Like, partly because I guess I really can't process

Travis Oliphant (2:57:07.480)

three screens at once anyway.

Lex Fridman (2:57:09.200)

Like, I just am looking at one and I just flip.

Travis Oliphant (2:57:12.560)

You know, I flip an application open.

Lex Fridman (2:57:14.460)

So where it's really helpful is actually

Travis Oliphant (2:57:17.340)

when I'm trying to do, you know,

Lex Fridman (2:57:18.440)

here's data and I want to input it from here.

Travis Oliphant (2:57:20.240)

Like this is the only time I really need another screen.

Lex Fridman (2:57:22.280)

So now, because you're both a developer, lead developers,

Lex Fridman (2:57:25.960)

but then there's also these businesses

Lex Fridman (2:57:27.880)

and there's salespeople and you're working

Travis Oliphant (2:57:30.120)

with large companies.

Lex Fridman (2:57:30.960)

Operations people, hiring people, yeah.

Travis Oliphant (2:57:32.480)

The whole thing.

Lex Fridman (2:57:33.400)

Which operating system is your favorite at this point?

Lex Fridman (2:57:37.240)

So Linux was the early days.

Lex Fridman (2:57:38.960)

So yeah, I love Linux as a server side.

Lex Fridman (2:57:41.460)

And it was early days I had my own Linux desktop.

Lex Fridman (2:57:44.340)

I've been on Mac laptops for 10 years now.

Travis Oliphant (2:57:47.800)

Yeah, this is what leadership looks like.

Lex Fridman (2:57:50.040)

As you switch to Mac.

Travis Oliphant (2:57:52.800)

Okay, great.

Lex Fridman (2:57:53.800)

Pretty much, I mean, just the fact that I had

Travis Oliphant (2:57:56.480)

to do PowerPoints, I had to do presentations

Lex Fridman (2:57:58.760)

and you know, plug in, I just couldn't mess

Travis Oliphant (2:58:01.240)

with plugging in laptops, it wouldn't project and yeah.

Lex Fridman (2:58:04.440)

So you mentioned also Quantset Labs and things like that.

Lex Fridman (2:58:09.240)

Can you give advice on how to hire great programmers

Lex Fridman (2:58:13.640)

and great people?

Travis Oliphant (2:58:14.600)

Yeah, I would say, produce an open source project,

Lex Fridman (2:58:19.400)

get people contributing to it and hire those people.

Travis Oliphant (2:58:21.560)

Yeah, I mean, you're doing it sort of,

Lex Fridman (2:58:25.080)

you may be perhaps a little biased,

Lex Fridman (2:58:27.080)

but that's probably 100% really good advice.

Lex Fridman (2:58:30.320)

I find it hard to hire.

Travis Oliphant (2:58:31.800)

I still find it hard to hire, like in terms of,

Lex Fridman (2:58:34.480)

I don't think that it's not hard to hire

Travis Oliphant (2:58:36.560)

if I've worked with somebody for a couple of weeks,

Lex Fridman (2:58:39.320)

but an hour or two of interviews, I have no idea.

Lex Fridman (2:58:43.600)

So that instinct, that radar of knowing if you're good

Lex Fridman (2:58:47.880)

or not, that you've found that you're still not able to.

Travis Oliphant (2:58:50.720)

It's really hard, I mean, the resume can help,

Lex Fridman (2:58:53.240)

but again, the resume is like a presentation

Travis Oliphant (2:58:55.520)

of the things they want you to see, not the reality of,

Lex Fridman (2:58:58.840)

and there's also, you have to understand

Lex Fridman (2:59:02.800)

what you're hiring for.

Lex Fridman (2:59:03.960)

There are different stages and different kinds of skills.

Lex Fridman (2:59:06.800)

And so it isn't just, one of the things I talk a lot about

Lex Fridman (2:59:10.940)

internally at my company is just that the whole idea

Travis Oliphant (2:59:14.440)

of measuring ourselves against a single axis is flawed

Lex Fridman (2:59:18.600)

because we're not, it's a multidimensional space

Lex Fridman (2:59:20.600)

and how do you order a multidimensional space?

Lex Fridman (2:59:22.120)

There isn't one ordering.

Lex Fridman (2:59:23.440)

So this whole idea, you immediately get projected

Lex Fridman (2:59:26.160)

into a thing when you're talking about hiring

Travis Oliphant (2:59:28.200)

or best or worst or better or not better.

Lex Fridman (2:59:30.660)

So what is the thing you're actually needing?

Lex Fridman (2:59:33.500)

And you can hire for that.

Lex Fridman (2:59:35.960)

There is such a thing, generally, I really value people

Travis Oliphant (2:59:39.040)

who have the affect, that care about open source.

Lex Fridman (2:59:42.920)

Like so in some cases, their affinity to open source

Travis Oliphant (2:59:45.720)

is simply kind of a filter of an affect.

Lex Fridman (2:59:49.100)

However, I have found this interesting dichotomy

Travis Oliphant (2:59:52.560)

between open source contributors and product creation.

Lex Fridman (2:59:58.520)

There's, I don't know if it's fully true,

Lex Fridman (30:00.540)

And they're related, those are two related questions.

Lex Fridman (30:02.540)

And then the debugging, like the iterative process

Travis Oliphant (30:05.500)

of running the script to figure out what the error is,

Lex Fridman (30:07.820)

maybe even for some people to do the fix yourself.

Lex Fridman (30:11.580)

So do you compile it?

Lex Fridman (30:12.660)

Do you, like how do you distribute that code to them?

Lex Fridman (30:15.620)

And it's interesting because I think

Lex Fridman (30:18.540)

it's exactly what you're talking about.

Travis Oliphant (30:20.100)

If you increase the circle of empathy,

Lex Fridman (30:24.260)

the circle of people that are able to use your programs,

Travis Oliphant (30:28.900)

you increase it, it's like effectiveness and it's power.

Lex Fridman (30:32.900)

And so you have to think, can I write scripts?

Travis Oliphant (30:37.020)

Can I write programs that can be used by medical engineers,

Lex Fridman (30:40.140)

by all kinds of people that don't know programming

Lex Fridman (30:43.900)

and actually maybe plant a seed,

Lex Fridman (30:46.900)

have them catch the bug of programming

Lex Fridman (30:48.380)

so that they start on a journey.

Lex Fridman (30:50.180)

That's a huge responsibility.

Lex Fridman (30:51.500)

And ultimately it has to do with the Amazon one click buy.

Lex Fridman (30:55.340)

Like how frictionless can you make the early steps?

Travis Oliphant (30:58.780)

Frictionless is actually really key.

Lex Fridman (31:00.380)

To go in any community is, any friction point,

Lex Fridman (31:03.020)

you're just gonna lose some people, right?

Lex Fridman (31:05.180)

Now sometimes you may wanna intentionally do that.

Travis Oliphant (31:09.060)

If you're early enough on, you need a lot of help.

Lex Fridman (31:11.620)

You need people who have the skills.

Travis Oliphant (31:13.340)

You might actually, it's helpful.

Lex Fridman (31:14.740)

You don't necessarily have too many users

Travis Oliphant (31:16.820)

as opposed to contributors if you're early on.

Lex Fridman (31:20.340)

Anyway, there's, SciFi started in 98,

Lex Fridman (31:23.100)

but it really emerged as this collection of modules

Lex Fridman (31:25.740)

that I was just putting on the net.

Lex Fridman (31:27.340)

People were downloading and I think I got 100 users, right?

Lex Fridman (31:31.580)

By the end of that year.

Lex Fridman (31:32.660)

But the fact that I got 100 users and more than that,

Lex Fridman (31:35.660)

people started to email me with fixes.

Lex Fridman (31:39.420)

And that was actually intoxicating, right?

Lex Fridman (31:41.300)

That was the, here I'm writing papers

Lex Fridman (31:44.220)

and I'm giving conferences and I get people to say hello,

Lex Fridman (31:46.180)

but yeah, good job.

Lex Fridman (31:47.420)

But mostly it was, you're viewed with,

Lex Fridman (31:49.860)

it's competitive, right?

Travis Oliphant (31:51.540)

You publish a paper and people are like,

Lex Fridman (31:52.900)

oh, it wasn't my paper.

Travis Oliphant (31:55.900)

I was starting to see that sense of academic life

Lex Fridman (31:59.220)

where it was so much,

Travis Oliphant (32:00.180)

I thought there was this cooperative effort,

Lex Fridman (32:01.460)

but it sounds like we're here just to one up each other.

Lex Fridman (32:04.940)

And it's not true across the board,

Lex Fridman (32:07.700)

but a lot of that's there.

Lex Fridman (32:08.580)

But here in this world,

Lex Fridman (32:09.660)

I was getting responses from people all over the world.

Lex Fridman (32:13.700)

I remember Pjaro Peterson in Estonia, right?

Lex Fridman (32:16.060)

Was one of the first people.

Lex Fridman (32:17.340)

And he sent me back this make file,

Lex Fridman (32:18.740)

cause the first thing it is, yeah, your build thing stinks

Lex Fridman (32:21.220)

and here's a better make file.

Lex Fridman (32:23.020)

Now it was a complex make file.

Travis Oliphant (32:24.380)

I don't think I never understood that make file actually,

Lex Fridman (32:26.580)

but it worked and it did a lot more.

Lex Fridman (32:29.220)

And so I said, thanks, this is cool.

Lex Fridman (32:30.980)

And that was my first kind of engagement

Travis Oliphant (32:32.500)

with community development.

Lex Fridman (32:35.100)

But the process was, he sent me a patch file.

Travis Oliphant (32:37.660)

I had to upload a new tar ball.

Lex Fridman (32:39.900)

And I just found, I really love that.

Lex Fridman (32:41.580)

And the style back then was here's a mailing list.

Lex Fridman (32:43.660)

It's very, it wasn't as,

Travis Oliphant (32:45.740)

it's certainly weren't the tools that are available today.

Lex Fridman (32:47.660)

It was very early on, but I really started to,

Travis Oliphant (32:49.940)

that's the whole year.

Lex Fridman (32:50.780)

I think I did about seven packages that year, right?

Lex Fridman (32:54.580)

And then by the end of the year,

Lex Fridman (32:55.540)

I collected them into a thing called multipack.

Lex Fridman (32:57.840)

So in 99, there was this thing called multipack.

Lex Fridman (32:59.780)

And that's when a high school student,

Travis Oliphant (33:01.820)

no, he was a high school student at the time,

Lex Fridman (33:03.060)

guy named Robert Kern,

Lex Fridman (33:04.780)

took that package and made a Windows installer, right?

Lex Fridman (33:09.700)

And then of course, a massive increase of usage.

Lex Fridman (33:12.700)

So by the way, most of this development was under Linux.

Lex Fridman (33:15.860)

Yes, yes, it was on Linux.

Travis Oliphant (33:17.380)

I was a Linux developer doing it on a Unix box.

Lex Fridman (33:20.240)

I mean, at the time I was actually getting into,

Travis Oliphant (33:23.020)

I had a new hard drive,

Lex Fridman (33:24.060)

did some kernel programming to make the hard drive work.

Travis Oliphant (33:26.500)

I mean, not programming, but modification to the kernel

Lex Fridman (33:28.780)

so I could actually get a hard drive working.

Travis Oliphant (33:31.180)

I love that aspect of it.

Lex Fridman (33:32.320)

I was also in, at school, I was building a cluster.

Travis Oliphant (33:36.100)

I took Mac computers and you put yellow dog Linux on them.

Lex Fridman (33:40.940)

At the Mayo Clinic, they were just,

Travis Oliphant (33:42.140)

they had all these Macs that were older,

Lex Fridman (33:43.520)

they were just getting rid of.

Lex Fridman (33:44.740)

And so I kind of got permission to go grab them together.

Lex Fridman (33:46.820)

I put about 24 of them together in a cluster, in a cabinet,

Lex Fridman (33:50.340)

and put yellow dog Linux on them all.

Lex Fridman (33:51.700)

And I wrote a C++ program to do MRI simulation.

Travis Oliphant (33:56.240)

That was what I was doing at the same time

Lex Fridman (33:58.900)

for my day job, so to speak.

Lex Fridman (34:01.400)

So I was loving the whole process.

Lex Fridman (34:03.460)

And the same time I was,

Travis Oliphant (34:04.300)

oh, I need a ordinary differential equation.

Lex Fridman (34:06.260)

That's why ordinary differential equations were key

Travis Oliphant (34:08.160)

was because that's the heart of a block equation

Lex Fridman (34:09.820)

for simulating MRI, is an ODE solver.

Lex Fridman (34:12.420)

And so that's, but I actually did that,

Lex Fridman (34:15.720)

it just happened at the same time.

Travis Oliphant (34:16.980)

That's why it was kind of what you're working on

Lex Fridman (34:18.540)

and what you're interested in, they're coinciding.

Travis Oliphant (34:20.500)

I was definitely scratching my own itch

Lex Fridman (34:22.380)

in terms of building stuff.

Lex Fridman (34:24.060)

And which helped in the sense that I was using it for me,

Lex Fridman (34:27.040)

so at least I had one user.

Travis Oliphant (34:28.540)

I had one person who was like, well, no, this is better.

Lex Fridman (34:30.360)

I like this interface better.

Lex Fridman (34:31.420)

And I had the experience of MATLAB

Lex Fridman (34:33.300)

to guide some of what those APIs might look like.

Lex Fridman (34:36.480)

But you're just doing yourself,

Lex Fridman (34:37.720)

you're building all this stuff.

Lex Fridman (34:39.000)

But with the Windows installer,

Lex Fridman (34:40.060)

it was the first time I realized, oh yeah,

Travis Oliphant (34:41.460)

the binary installer really helps people.

Lex Fridman (34:43.740)

And so that led to spending more time

Travis Oliphant (34:46.980)

on that side of things.

Lex Fridman (34:49.100)

So around 2000, so I graduated my PhD in 2000,

Travis Oliphant (34:52.780)

end of year, end of 2000.

Lex Fridman (34:53.780)

So 99 doing a lot of work there,

Travis Oliphant (34:56.660)

98 doing a lot of work there,

Lex Fridman (34:57.740)

99 kind of spending more time on my PhD,

Travis Oliphant (35:00.780)

helping people use the tools,

Lex Fridman (35:02.420)

thinking about what do I want to go from here.

Travis Oliphant (35:04.060)

There was a company, there was a guy actually,

Lex Fridman (35:05.620)

Eric Jones and Travis Vought.

Travis Oliphant (35:07.620)

They were two friends who founded a company called NTHOT.

Lex Fridman (35:11.080)

It's here in Austin, still here.

Lex Fridman (35:13.620)

And they, Eric contacted me at the time

Lex Fridman (35:16.060)

when I was a graduate student still.

Lex Fridman (35:19.380)

And he said, hey, why don't you come down?

Lex Fridman (35:20.860)

We want to build a company.

Travis Oliphant (35:22.660)

We're thinking of a scientific company

Lex Fridman (35:25.720)

and we want to take what you're doing

Lex Fridman (35:27.560)

and kind of add it to some stuff that he'd done.

Lex Fridman (35:29.460)

He'd written some tools.

Lex Fridman (35:31.220)

And then Piero Peterson had done F2Py.

Lex Fridman (35:32.820)

Let's come together and build,

Travis Oliphant (35:34.380)

pull this all together and call it SciPy.

Lex Fridman (35:36.740)

So that's the origin of the SciPy brand.

Travis Oliphant (35:39.480)

It came from multi pack

Lex Fridman (35:41.380)

and a whole bunch of modules I'd written,

Travis Oliphant (35:42.580)

plus a few things from some other folks

Lex Fridman (35:44.500)

and then pulled together in a single installer.

Travis Oliphant (35:47.580)

SciPy was really a distribution of Python

Lex Fridman (35:49.540)

masquerading as a library.

Lex Fridman (35:51.260)

How did you think about SciPy in context of Python,

Lex Fridman (35:54.340)

in context of Numeric, like what?

Lex Fridman (35:56.180)

So we saw SciPy as a way to make an R&D environment

Lex Fridman (35:59.020)

for Python, like use Python, depended on Numeric.

Lex Fridman (36:03.380)

So Numeric was the array library we depended on.

Lex Fridman (36:05.540)

And then from there, extend it with a bunch of modules

Travis Oliphant (36:08.260)

that allowed for, and at the time,

Lex Fridman (36:10.340)

the original vision of SciPy was to have plotting,

Travis Oliphant (36:13.180)

was to have the REPL environment

Lex Fridman (36:16.140)

and kind of really a whole data environment

Travis Oliphant (36:19.500)

that you could then install and get going with.

Lex Fridman (36:21.020)

And that was kind of the thinking.

Lex Fridman (36:23.020)

It didn't really evolve that way, right?

Lex Fridman (36:25.020)

It sort of had a, for one,

Travis Oliphant (36:27.580)

it's really hard to do massive scale projects

Lex Fridman (36:31.940)

with open source collectives.

Travis Oliphant (36:34.300)

Actually, there's sort of an intrinsic cooperation limit

Lex Fridman (36:38.500)

as to which, too many cooks in the kitchen,

Travis Oliphant (36:40.780)

you can do amazing infrastructure work.

Lex Fridman (36:42.780)

When it comes down to bringing it all together

Travis Oliphant (36:44.220)

into a single deliverable,

Lex Fridman (36:45.860)

that actually requires a little more product management

Travis Oliphant (36:49.660)

that is not, that doesn't really emerge

Lex Fridman (36:52.820)

from the same dynamic.

Lex Fridman (36:53.980)

So it struggled, struggled to get almost too many voices.

Lex Fridman (36:57.860)

It's hard to have everybody agree.

Travis Oliphant (36:59.220)

Consensus doesn't really work at that scale.

Lex Fridman (37:02.100)

You end up with politics,

Travis Oliphant (37:03.260)

with the same kind of things that's happened

Lex Fridman (37:05.220)

in large organizations trying to decide

Lex Fridman (37:07.100)

what to do together.

Lex Fridman (37:09.380)

So consensus building was challenging at scale

Lex Fridman (37:12.340)

as more people came in, right?

Lex Fridman (37:13.860)

Early on, it's fine, because there's nobody there.

Lex Fridman (37:15.700)

So it works, but then as you get more successful

Lex Fridman (37:17.740)

and more people use it, all of a sudden,

Travis Oliphant (37:18.980)

oh, there's this scale at which this doesn't work anymore

Lex Fridman (37:22.300)

and we have to come up with different approaches.

Lex Fridman (37:23.980)

So Sidepy came out officially in 2001,

Lex Fridman (37:26.700)

was the first release, most of the time.

Travis Oliphant (37:28.900)

I remember the days of getting that release ready.

Lex Fridman (37:31.060)

It was a Windows installer and there were bugs

Travis Oliphant (37:33.420)

on how the Windows compiler handled complex numbers

Lex Fridman (37:36.300)

and you were chasing segmentation faults.

Lex Fridman (37:38.540)

And it was, it's a lot of work.

Lex Fridman (37:40.420)

There was a lot of effort had nothing to do

Travis Oliphant (37:43.140)

with my area of study.

Lex Fridman (37:45.540)

And at the same time, I had just gotten an offer.

Lex Fridman (37:47.500)

So he wondered if I wanted to come down

Lex Fridman (37:48.780)

and help him start that company with his friend.

Lex Fridman (37:51.460)

And at the time I was like, I was intrigued,

Lex Fridman (37:53.380)

but I was squaring a path, an academic path.

Lex Fridman (37:56.620)

And I had just got an offer to go and teach at my alma mater.

Lex Fridman (37:59.980)

So I took that tenure track position.

Lex Fridman (38:02.420)

And Sidepy, and kind of, then I started to work on Sidepy

Lex Fridman (38:05.180)

as a professor too.

Lex Fridman (38:07.060)

So that's, I left, I've got the Mayo Clinic,

Lex Fridman (38:09.540)

graduated, wrote my thesis using Sidepy,

Travis Oliphant (38:11.700)

wrote, you know, there's images that were created.

Lex Fridman (38:15.500)

Now the plotting tool I used was something

Travis Oliphant (38:17.300)

from Yorick actually.

Lex Fridman (38:18.660)

It was a plotting, a PLT kind of a plotting language

Travis Oliphant (38:21.940)

that I used.

Lex Fridman (38:22.780)

Yorick is a programming language?

Travis Oliphant (38:23.940)

It was a programming language, had a plotting tool,

Lex Fridman (38:26.340)

Dyslin, it had integration to Dyslin.

Travis Oliphant (38:28.940)

I ended up using Dyslin plus some of the plotting

Lex Fridman (38:31.340)

from Yorick linked to from Python.

Travis Oliphant (38:33.740)

Anyway, it was, people don't plot that way now,

Lex Fridman (38:37.180)

but this is before, and Sidepy was trying to add plotting.

Lex Fridman (38:40.260)

Yeah. Right?

Lex Fridman (38:41.460)

It didn't have much success.

Travis Oliphant (38:42.580)

Really the success of plotting came from John Hunter,

Lex Fridman (38:45.580)

who had a similar experience to my experience,

Travis Oliphant (38:47.420)

my kind of maverick experience as a person

Lex Fridman (38:49.660)

just trying to get stuff done and kind of having more time

Lex Fridman (38:51.700)

than money maybe, right?

Lex Fridman (38:53.820)

And John Hunter created what?

Travis Oliphant (38:55.300)

MapPlotLib.

Lex Fridman (38:56.300)

He's the creator of MapPlotLib.

Travis Oliphant (38:57.140)

Yeah, so John Hunter was, you know,

Lex Fridman (38:59.140)

he wasn't a student at the time, but he was an,

Travis Oliphant (39:00.580)

he was working in Quant field and he said,

Lex Fridman (39:02.120)

we need better plotting.

Lex Fridman (39:03.500)

So he just went out and said, cool, I'll make a new project

Lex Fridman (39:05.540)

and we'll call it MapPlotLib.

Lex Fridman (39:06.580)

And he released in 2001,

Lex Fridman (39:08.260)

about the same time that Sidepy came out

Lex Fridman (39:09.920)

and it was separate library, separate install,

Lex Fridman (39:12.960)

use numeric, Sidepy use numeric.

Lex Fridman (39:15.540)

And so Sidepy, you know, in 2001, we released Sidepy

Lex Fridman (39:18.980)

and then Endthought created a conference called Sidepy,

Travis Oliphant (39:22.380)

which was brought people together to talk about the space.

Lex Fridman (39:25.460)

And that conference is still ongoing.

Travis Oliphant (39:26.700)

It's one of the favorite conferences of a lot of people

Lex Fridman (39:28.460)

because it's, you know, it's changed over the years,

Lex Fridman (39:30.820)

but early on it was, you know, a collection of 50 people

Lex Fridman (39:33.740)

who care about, scientists mostly, you know,

Travis Oliphant (39:36.700)

practicing scientists who want, who care about coding

Lex Fridman (39:39.300)

and doing it well and not using MATLAB.

Lex Fridman (39:42.140)

And I remember being driven by, you know, I liked MATLAB,

Lex Fridman (39:44.120)

but I didn't like the fact that,

Lex Fridman (39:46.420)

so I'm not opposed to proprietary software.

Lex Fridman (39:48.060)

I'm actually not an open source zealot.

Travis Oliphant (39:50.220)

I love open source for the, what it brings,

Lex Fridman (39:52.660)

but I also see the role for proprietary software.

Lex Fridman (39:54.460)

But what I didn't like was the fact that I would develop

Lex Fridman (39:56.580)

code and publish it and then effectively telling somebody

Travis Oliphant (39:59.940)

here to run my code, you have to have

Lex Fridman (3:00:00.560)

but there does seem to be the more experienced,

Travis Oliphant (3:00:04.960)

the more affect somebody has an open source community,

Lex Fridman (3:00:08.160)

the less ability to actually produce product that they have.

Lex Fridman (3:00:11.640)

And the opposite is kind of true too.

Lex Fridman (3:00:13.520)

The more product focused are, I find a lot of people,

Travis Oliphant (3:00:16.000)

I've talked to a lot of people who produce

Lex Fridman (3:00:17.020)

really great products and they have a,

Travis Oliphant (3:00:19.400)

they're looking over the open source communities,

Lex Fridman (3:00:21.120)

kind of wanting to participate and play,

Lex Fridman (3:00:23.320)

but they've played here and they do a great job here

Lex Fridman (3:00:26.000)

and then they don't necessarily have some of the same.

Travis Oliphant (3:00:29.520)

Now I don't think that's entirely necessary.

Lex Fridman (3:00:32.040)

I think part of it is cultural, how they've emerged.

Travis Oliphant (3:00:34.880)

Because one of the things that open source communities

Lex Fridman (3:00:36.300)

often lack is great product management,

Travis Oliphant (3:00:39.160)

like some product management energy.

Lex Fridman (3:00:41.000)

That's brilliant, but you want both of those energies

Travis Oliphant (3:00:43.600)

in the same place together.

Lex Fridman (3:00:44.880)

Yes, you really do.

Lex Fridman (3:00:45.840)

And so a lot of it's creating these teams of people

Lex Fridman (3:00:48.120)

that have these needed skills and attributes

Travis Oliphant (3:00:50.480)

that are hard.

Lex Fridman (3:00:51.880)

And so one of the big things I look for is somebody

Travis Oliphant (3:00:55.120)

that fundamentally recognizes their need to learn.

Lex Fridman (3:00:57.800)

Like one of the values that we have

Travis Oliphant (3:00:59.560)

in all of the things we do is learning.

Lex Fridman (3:01:01.400)

Like if somebody thinks they know it all,

Travis Oliphant (3:01:04.560)

they're gonna struggle.

Lex Fridman (3:01:06.240)

And some of that is just, there's more basic things

Travis Oliphant (3:01:09.440)

like humility, just being humble in the face

Lex Fridman (3:01:12.780)

of all the things you don't know.

Lex Fridman (3:01:14.400)

And that's step one of learning.

Lex Fridman (3:01:15.840)

That's step one of learning, right?

Lex Fridman (3:01:16.960)

And I've spent a lot of time learning, right?

Lex Fridman (3:01:20.840)

Other people spend a lot more time,

Lex Fridman (3:01:21.840)

but I've spent a lot of time learning.

Lex Fridman (3:01:23.280)

My whole goal was to get a PhD because I love school

Lex Fridman (3:01:26.320)

and I wanted to be a scientist.

Lex Fridman (3:01:28.240)

And then what I found is what's been written about

Travis Oliphant (3:01:31.120)

elsewhere as well is the more I learned,

Lex Fridman (3:01:32.600)

the more I didn't know.

Travis Oliphant (3:01:33.780)

The more I realized, man, I know about this,

Lex Fridman (3:01:37.680)

but this is such a tiny thing in the global scope

Travis Oliphant (3:01:40.060)

of what I might wanna know about.

Lex Fridman (3:01:41.220)

So I need to be listening a whole lot better

Travis Oliphant (3:01:43.840)

than I am just talking.

Lex Fridman (3:01:47.360)

That's changed a little bit actually.

Travis Oliphant (3:01:48.840)

My wife says that I used to be a better listener.

Lex Fridman (3:01:50.600)

Now that I'm so full of all these ideas I wanna do,

Travis Oliphant (3:01:52.880)

she kind of says, you gotta give people time to talk.

Lex Fridman (3:01:55.520)

So you've succeeded on multiple dimensions.

Lex Fridman (3:01:58.400)

So one is the tenure track faculty.

Lex Fridman (3:02:01.680)

The other is just creating all these products

Lex Fridman (3:02:03.080)

and building up the businesses,

Lex Fridman (3:02:04.320)

then working with businesses.

Lex Fridman (3:02:06.880)

Do you have advice for young people today

Lex Fridman (3:02:09.240)

in high school and college of how to live a life

Travis Oliphant (3:02:13.880)

as nonlinear and as successful as yours,

Lex Fridman (3:02:18.280)

a life that they could be proud of?

Travis Oliphant (3:02:21.200)

Well, that's a super compliment.

Lex Fridman (3:02:22.960)

I'm humbled by that actually.

Travis Oliphant (3:02:24.200)

I would say a life they can be proud of.

Lex Fridman (3:02:27.960)

Honestly, one thing that I've said to people is first,

Travis Oliphant (3:02:31.560)

find people you love and care about them.

Lex Fridman (3:02:34.240)

Like family matters to me a lot.

Lex Fridman (3:02:36.040)

And family means people you love and have committed to.

Lex Fridman (3:02:39.640)

So it can be whatever you mean by that,

Lex Fridman (3:02:42.160)

but you need to have a foundation.

Lex Fridman (3:02:45.120)

So find people you love and wanna commit to and do that.

Travis Oliphant (3:02:48.960)

Cause it anchors you in a way that nothing else can.

Lex Fridman (3:02:52.200)

And then you find other things.

Lex Fridman (3:02:55.200)

And then kind of from out there,

Lex Fridman (3:02:56.640)

you find other kinds of things you can commit to,

Travis Oliphant (3:02:58.800)

whether it's ideas or people or groups of people.

Lex Fridman (3:03:03.240)

So, especially in high school,

Travis Oliphant (3:03:06.400)

I would say don't settle on what you think you know.

Lex Fridman (3:03:09.320)

Like give yourself 10 years to think about the world.

Travis Oliphant (3:03:13.320)

Like I see a lot of high school students

Lex Fridman (3:03:15.440)

who seem to know everything already.

Travis Oliphant (3:03:17.640)

I think I did too.

Lex Fridman (3:03:18.720)

I think it's maybe natural,

Lex Fridman (3:03:20.360)

but recognize that the things you care about,

Lex Fridman (3:03:23.160)

you might change your perspective over time.

Travis Oliphant (3:03:26.520)

I certainly have over time.

Lex Fridman (3:03:28.600)

I was really passionate about one specific thing

Lex Fridman (3:03:30.640)

and I was kind of softened.

Lex Fridman (3:03:32.520)

I was a big, I didn't like the Federal Reserve, right?

Lex Fridman (3:03:35.760)

And there's still, we could have a longer conversation

Lex Fridman (3:03:38.480)

about monetary policy and finances,

Lex Fridman (3:03:40.120)

but I'm a little more nuanced in my perspective

Lex Fridman (3:03:46.000)

at this point.

Lex Fridman (3:03:48.000)

But that's one area where you learn about something,

Lex Fridman (3:03:50.160)

go, ah, I wanna attack it.

Travis Oliphant (3:03:52.440)

Build, don't destroy.

Lex Fridman (3:03:55.160)

Build, like so often the tendency is to not like something

Lex Fridman (3:03:58.400)

and wanna go attack it.

Lex Fridman (3:04:00.000)

Build something, build something to replace it.

Travis Oliphant (3:04:02.240)

Yeah.

Lex Fridman (3:04:03.080)

Build up, attract people to your new thing.

Lex Fridman (3:04:05.600)

You'll be far better, right?

Lex Fridman (3:04:08.800)

You don't need to destroy something to build something else.

Lex Fridman (3:04:12.600)

So that's, I guess, generally.

Lex Fridman (3:04:14.880)

And then definitely like curiosity,

Travis Oliphant (3:04:19.120)

follow your curiosity and let it,

Lex Fridman (3:04:22.680)

don't just follow the money.

Lex Fridman (3:04:24.600)

And all of that, like you said,

Lex Fridman (3:04:25.800)

is grounded in family, friendship, and ultimately love.

Travis Oliphant (3:04:30.160)

Yes.

Lex Fridman (3:04:31.200)

Which is a great way to end it.

Travis Oliphant (3:04:34.640)

Travis, you're one of the most impactful people

Lex Fridman (3:04:37.080)

in the engineering and the computer science

Travis Oliphant (3:04:38.760)

in the human world.

Lex Fridman (3:04:39.920)

So I truly appreciate everything you've done.

Lex Fridman (3:04:43.520)

And I really appreciate that you would spend

Lex Fridman (3:04:45.800)

your valuable time with me.

Travis Oliphant (3:04:46.960)

It was an honor.

Lex Fridman (3:04:47.800)

It was a real pleasure for me.

Travis Oliphant (3:04:48.840)

I appreciate that.

Lex Fridman (3:04:50.520)

Thanks for listening to this conversation

Travis Oliphant (3:04:52.080)

with Travis Oliphant.

Lex Fridman (3:04:54.000)

To support this podcast,

Travis Oliphant (3:04:55.320)

please check out our sponsors in the description.

Lex Fridman (3:04:57.900)

And now, let me leave you with something

Travis Oliphant (3:05:00.200)

that in the programming world is called Hodgson's Law.

Lex Fridman (3:05:04.960)

Every sufficiently advanced Lisp application

Travis Oliphant (3:05:08.120)

will eventually be re implemented in Python.

Lex Fridman (3:05:12.520)

Thank you for listening and hope to see you next time.

Travis Oliphant (40:01.420)

this proprietary software.

Lex Fridman (40:02.500)

Right, and there's also culture around MATLAB as much,

Travis Oliphant (40:05.940)

because I've talked to a few folks in,

Lex Fridman (40:08.260)

MathWorks creates MATLAB?

Travis Oliphant (40:09.820)

Yeah.

Lex Fridman (40:10.820)

I mean, there's just a culture, they try really hard,

Lex Fridman (40:13.900)

but it just, there's this corporate IBM style culture

Lex Fridman (40:16.820)

that's like, or whatever.

Travis Oliphant (40:18.380)

I don't want to say negative things about IBM or whatever,

Lex Fridman (40:20.780)

but there's a...

Travis Oliphant (40:22.260)

No, it's really that connection.

Lex Fridman (40:23.740)

It's something I'm in the middle of right now

Travis Oliphant (40:24.940)

is the business of open source.

Lex Fridman (40:27.000)

And how do you connect the ethos of cooperative development

Lex Fridman (40:30.820)

with the necessity of creating profits, right?

Lex Fridman (40:34.780)

And like right now today, I'm still in the middle of that.

Travis Oliphant (40:38.060)

That's actually the early days of me exploring this question.

Lex Fridman (40:42.260)

Cause I was writing SciPy, I mean, as an aside,

Travis Oliphant (40:44.660)

I also had, so I had three kids at the time.

Lex Fridman (40:46.540)

I have six kids now.

Travis Oliphant (40:47.860)

I got married early, wanted a family.

Lex Fridman (40:50.860)

I had three kids and I remember reading,

Travis Oliphant (40:52.620)

I read Richard Stallman's post and I was a fan of Stallman.

Lex Fridman (40:55.540)

I would read his work, I liked this collective ideas

Travis Oliphant (40:58.100)

he would have.

Lex Fridman (40:58.940)

Certainly the ideas on IP law, I read a lot of his stuff.

Lex Fridman (41:01.740)

But then he said, okay, well,

Lex Fridman (41:04.820)

how do I make money with this?

Lex Fridman (41:05.780)

How do I make a living?

Lex Fridman (41:06.700)

How do I pay for my kids?

Travis Oliphant (41:07.740)

All this stuff was in my mind,

Lex Fridman (41:09.300)

young graduate student making no money,

Travis Oliphant (41:10.640)

thinking I got to get a job.

Lex Fridman (41:12.060)

And he said, well, I think just be like me

Lex Fridman (41:14.540)

and don't have kids, right?

Lex Fridman (41:15.840)

That's just, don't, don't.

Travis Oliphant (41:17.080)

That's his take on it.

Lex Fridman (41:18.540)

That was what he said in that moment, right?

Travis Oliphant (41:20.860)

That's the thing I read and I went,

Lex Fridman (41:22.420)

okay, this is a train I can't get on.

Travis Oliphant (41:24.960)

There has to be a way to preserve the culture

Lex Fridman (41:26.700)

of open source and still be able to make sufficient money

Travis Oliphant (41:29.180)

to feed your kids.

Lex Fridman (41:30.020)

Yes, exactly, there's gotta be.

Travis Oliphant (41:31.500)

Well, so that actually led me to a study of economics.

Lex Fridman (41:34.500)

Because at the time I was ignorant and I really was.

Travis Oliphant (41:36.680)

I'm actually, I'm embarrassed for educational system

Lex Fridman (41:39.420)

that they could let me and I was valedictorian

Travis Oliphant (41:41.300)

in my high school class and I did super well in college.

Lex Fridman (41:43.720)

And like academically I did great, right?

Lex Fridman (41:47.620)

But the fact that I could do that and then be clueless

Lex Fridman (41:49.980)

about this key part of life,

Travis Oliphant (41:52.740)

it led me to go, there's a problem.

Lex Fridman (41:54.400)

Like I should have learned this in fifth grade.

Travis Oliphant (41:56.660)

I should have learned this in eighth grade.

Lex Fridman (41:58.380)

Like everybody should come out

Travis Oliphant (41:59.220)

with a basic knowledge of economics.

Lex Fridman (42:01.700)

You're an interesting example because you've created tools

Travis Oliphant (42:04.040)

that change the lives of probably millions of people

Lex Fridman (42:07.640)

and the fact that you don't understand at the time

Travis Oliphant (42:10.060)

of the creation of those tools, the basics economics

Lex Fridman (42:12.860)

of how like to build up a giant system is the problem.

Travis Oliphant (42:15.260)

Yeah, it's a problem.

Lex Fridman (42:16.100)

And so during my PhD at the same time,

Travis Oliphant (42:18.260)

this is back in 98, 99 at the same time,

Lex Fridman (42:20.720)

I was in a library, I was reading books on capitalism,

Travis Oliphant (42:23.380)

I was reading books on Marxism,

Lex Fridman (42:24.700)

I was reading books on what is this thing?

Lex Fridman (42:27.700)

What does it mean?

Lex Fridman (42:29.700)

And I encountered, basically I encountered a set of writings

Travis Oliphant (42:33.140)

from people that said they were the inheritors of Adam Smith.

Lex Fridman (42:35.500)

Read Adam Smith for the first time, right?

Travis Oliphant (42:37.220)

Which is the wealth of nations

Lex Fridman (42:38.580)

and kind of this notion of emergent societies

Lex Fridman (42:42.460)

and realized, oh, there's this whole world out here

Lex Fridman (42:45.100)

of people and the challenge of economics is also political.

Travis Oliphant (42:49.460)

Like, cause economics, people, different parties

Lex Fridman (42:53.940)

running for office, they want their economic friends.

Lex Fridman (42:58.080)

They want their economists to back them up, right?

Lex Fridman (43:00.040)

Or to be their magicians, like the magicians

Lex Fridman (43:03.700)

in Pharaoh's court, right?

Lex Fridman (43:04.660)

The people that are kind of say, hey, this is,

Travis Oliphant (43:06.260)

you should listen to me because I've got the expert

Lex Fridman (43:08.100)

who says this.

Lex Fridman (43:09.420)

And so it gets really muddled, right?

Lex Fridman (43:11.540)

But I was looking at it from as a scientist going,

Lex Fridman (43:14.020)

what is this space?

Lex Fridman (43:14.860)

What does this mean?

Lex Fridman (43:15.680)

How does Paris get fed?

Lex Fridman (43:16.940)

How does, what is money?

Lex Fridman (43:18.420)

How does it work?

Lex Fridman (43:19.420)

And I found a lot of writings that I really loved.

Travis Oliphant (43:21.580)

I found some things that I really loved

Lex Fridman (43:22.860)

and I learned from that.

Travis Oliphant (43:23.980)

It was writings from people like Von Missess.

Lex Fridman (43:26.300)

He wrote a paper in 1920 that still should be read

Travis Oliphant (43:29.060)

more than it is.

Lex Fridman (43:29.900)

It was the economic calculation problem

Travis Oliphant (43:33.060)

of the socialist commonwealth.

Lex Fridman (43:34.560)

It was basically in response

Travis Oliphant (43:35.420)

to the Bolshevik revolution in 1917.

Lex Fridman (43:37.140)

And his basic argument was it's not gonna work

Travis Oliphant (43:40.180)

to not have private property.

Lex Fridman (43:41.780)

You're not gonna be able to come up with prices.

Travis Oliphant (43:43.420)

The bureaucrats aren't gonna be able to determine

Lex Fridman (43:45.200)

how to allocate resources without a price system.

Lex Fridman (43:47.620)

And a price system emerges from people making trades.

Lex Fridman (43:51.700)

And they can only make trades if they have authority

Travis Oliphant (43:53.860)

over the thing they're trading.

Lex Fridman (43:55.460)

And that creates information flow

Travis Oliphant (43:58.020)

that you just don't have if you try to top down it.

Lex Fridman (44:01.300)

Right.

Lex Fridman (44:02.140)

And it's like, huh, that's a really good point.

Lex Fridman (44:04.780)

Yeah, the prices have a signal that's used.

Lex Fridman (44:06.860)

And it's important to have that signal

Lex Fridman (44:09.400)

when you're trying to build a community

Travis Oliphant (44:11.020)

of productive people like you would

Lex Fridman (44:12.580)

in the software engineering space.

Travis Oliphant (44:13.700)

Yeah, the prices are actually

Lex Fridman (44:14.860)

an important signaling mechanism.

Travis Oliphant (44:17.540)

Right, and that money is just a bartering tool.

Lex Fridman (44:20.820)

Right, so this is the first time I've encountered

Travis Oliphant (44:22.540)

any of this concept, right, and the fact that,

Lex Fridman (44:24.440)

oh, this is actually really critical.

Travis Oliphant (44:26.600)

Like it's so critical to our prosperity

Lex Fridman (44:29.340)

and that we're dangerously not learning about this,

Travis Oliphant (44:34.100)

not teaching our children about this.

Lex Fridman (44:36.140)

So you had the three kids,

Travis Oliphant (44:37.260)

you had to make some hard decisions.

Lex Fridman (44:38.080)

I had to make some money, right, had to figure it out.

Lex Fridman (44:39.880)

But I didn't really care.

Lex Fridman (44:40.720)

I mean, I've never been driven by money, just need it.

Travis Oliphant (44:43.260)

Yeah, right, need to eat.

Lex Fridman (44:45.200)

So how did that resolve itself in terms of site buy?

Lex Fridman (44:49.100)

So I would say it didn't really resolve itself.

Lex Fridman (44:51.320)

It sort of started a journey that I'm continuing on.

Travis Oliphant (44:53.420)

I'm still on, I would say.

Lex Fridman (44:54.740)

I don't think it resolved itself.

Lex Fridman (44:55.660)

But I will say I went in eyes wide open.

Lex Fridman (44:59.260)

Like I knew that there were problems

Travis Oliphant (45:00.940)

with giving stuff away and creating the market externalities

Lex Fridman (45:07.900)

that the fact that, yeah, people might use it

Lex Fridman (45:09.780)

and I might not get paid for it

Lex Fridman (45:10.820)

and I'll have to figure something else out to get paid.

Travis Oliphant (45:13.060)

Like at least I can say I'm not bitter

Lex Fridman (45:14.940)

that a lot of people have used stuff that I've written

Lex Fridman (45:17.220)

and I haven't necessarily benefited economically from it.

Lex Fridman (45:20.240)

I've heard other people be bitter about that

Travis Oliphant (45:22.300)

when they write or they talk.

Lex Fridman (45:23.300)

Like, oh, I should've got more value out of this.

Lex Fridman (45:24.900)

And I'm also, I want to create systems

Lex Fridman (45:27.740)

that let people like me who might have these desires

Travis Oliphant (45:31.060)

to do things, let them benefit.

Lex Fridman (45:32.260)

So it actually creates more of the same.

Travis Oliphant (45:34.700)

Not to turn on your bitterness module,

Lex Fridman (45:36.900)

but there's some aspect, I wish there was mechanisms for me

Travis Oliphant (45:40.940)

to reward whoever created side buy and non buy

Lex Fridman (45:43.580)

because it brought so much joy to my life.

Travis Oliphant (45:45.300)

I appreciate that.

Lex Fridman (45:46.140)

You know what I mean?

Travis Oliphant (45:46.980)

The tip dark notion was there.

Lex Fridman (45:48.340)

I appreciate that.

Lex Fridman (45:49.180)

But there should be a very frictionless mechanism.

Lex Fridman (45:51.940)

There should be a frictionless mechanism.

Travis Oliphant (45:52.760)

I totally agree.

Lex Fridman (45:53.600)

I would love to talk about some of the ideas I have

Travis Oliphant (45:55.220)

because I actually came across,

Lex Fridman (45:56.220)

I think I've come up with some interesting notions

Travis Oliphant (45:58.200)

that could work, but they'll require anything that will work

Lex Fridman (46:01.860)

takes time to emerge, right?

Travis Oliphant (46:03.740)

Like things don't just turn overnight.

Lex Fridman (46:04.940)

That's definitely one thing I've also understood

Lex Fridman (46:06.340)

and learned is any fixes, that's why it's kind of funny.

Lex Fridman (46:10.120)

We often give credit to, oh, this president gets elected

Lex Fridman (46:12.940)

and oh, look how great things have done.

Lex Fridman (46:14.420)

And I saw that when I had a transition in a condo

Lex Fridman (46:18.340)

when a new CEO came in, right?

Lex Fridman (46:19.520)

And it's like the success that's happening,

Travis Oliphant (46:22.340)

there's an inertia there.

Lex Fridman (46:23.460)

Yeah, and sometimes the decision you made

Travis Oliphant (46:25.740)

like 10 years before is the reason why the success is the.

Lex Fridman (46:28.980)

Right, exactly.

Lex Fridman (46:29.820)

So we're sort of just running around taking credit

Lex Fridman (46:31.560)

for stuff.

Travis Oliphant (46:32.400)

The credit assignment has like a delay to it

Lex Fridman (46:35.140)

that makes the credit assignment basically wrong

Travis Oliphant (46:38.320)

more than right.

Lex Fridman (46:39.160)

Wrong more than right, exactly.

Lex Fridman (46:40.320)

And so I'm like, oh, this is, you know,

Lex Fridman (46:42.140)

that's the stuff I would read a ton about, you know,

Travis Oliphant (46:44.860)

early on.

Lex Fridman (46:45.700)

So I don't, I feel like I'm with you.

Travis Oliphant (46:47.720)

Like I want the same thing.

Lex Fridman (46:48.780)

I want to be able to, and honestly, not for personally,

Travis Oliphant (46:50.900)

I've been happy.

Lex Fridman (46:51.740)

I've been happy.

Travis Oliphant (46:52.720)

I feel like I don't have any, I mean,

Lex Fridman (46:53.980)

we've been done reasonably okay, but I've had to pursue it.

Travis Oliphant (46:56.920)

Like that's really what started my trajectory from academia

Lex Fridman (47:01.380)

is reading that stuff led me to say,

Travis Oliphant (47:02.940)

oh, entrepreneurship matters.

Lex Fridman (47:05.780)

So I love software, but we need more entrepreneurs

Lex Fridman (47:09.180)

and I wanna understand that better.

Lex Fridman (47:10.360)

So once I kind of had that virus infect my brain,

Travis Oliphant (47:16.500)

even though I was on a trajectory

Lex Fridman (47:17.580)

to go to a tenure track position at a university

Lex Fridman (47:20.640)

and I was there for six years,

Lex Fridman (47:22.780)

I was kind of already out the door when I started.

Lex Fridman (47:26.060)

And we can get into that, but.

Lex Fridman (47:27.660)

Well, can I just ask you a quick question on,

Travis Oliphant (47:30.340)

is there some design principles

Lex Fridman (47:32.740)

that were in your mind around SciPy?

Travis Oliphant (47:34.740)

Like, is there some key ideas

Lex Fridman (47:36.460)

that were just like sticking to you

Lex Fridman (47:38.060)

that this is the fundamental ideas?

Lex Fridman (47:40.300)

Yeah, I would say so.

Travis Oliphant (47:41.140)

I would think it's basically accessibility to scientists,

Lex Fridman (47:43.680)

like give them, give scientists and engineers tools

Travis Oliphant (47:46.980)

that they don't have to think a lot about programming.

Lex Fridman (47:48.380)

So give them really good building blocks,

Travis Oliphant (47:50.300)

give them functions that they wanna call

Lex Fridman (47:51.860)

and sort of just the right length of spelling.

Travis Oliphant (47:55.860)

There's one tradition in programming where it's like,

Lex Fridman (47:59.500)

make very, very long names, right?

Lex Fridman (48:01.880)

And you can see it in some programming languages

Lex Fridman (48:03.700)

where the names get, take half the screen.

Lex Fridman (48:06.460)

And in the 4chan world, characters had to be six letters

Lex Fridman (48:11.540)

early on, right?

Lex Fridman (48:12.380)

And that's way too much, too little.

Lex Fridman (48:14.340)

But I was like, I liked to have names

Travis Oliphant (48:16.820)

that were informative but short.

Lex Fridman (48:18.940)

So even though Python, well this is a different conversation,

Lex Fridman (48:22.340)

but documentation is doing some work there.

Lex Fridman (48:25.860)

So when you look at great scientific libraries

Lex Fridman (48:29.180)

and functions, there's a richness of documentation

Lex Fridman (48:32.700)

that helps you get into the details.

Travis Oliphant (48:34.820)

The first glance at a function gives you the intuition

Lex Fridman (48:37.620)

of all it needs to do by looking at the headers and so on.

Lex Fridman (48:40.540)

But to get the depths of all the complexities involved,

Lex Fridman (48:43.420)

all the options involved,

Travis Oliphant (48:44.740)

documentation does some of the work.

Lex Fridman (48:45.580)

Documentation is essential, yeah.

Lex Fridman (48:47.380)

So that was actually a, so we thought about several things.

Lex Fridman (48:50.520)

One is we wanted plotting.

Travis Oliphant (48:51.940)

We wanted interactive environment.

Lex Fridman (48:53.580)

We wanted good documentation.

Travis Oliphant (48:54.860)

These are things we knew, we wanted.

Lex Fridman (48:56.780)

The reality is those took about 10 years to evolve, right?

Travis Oliphant (49:00.460)

Given the fact that we didn't have a big budget,

Lex Fridman (49:02.060)

it was all volunteer labor.

Travis Oliphant (49:03.100)

It was sort of, when nthought got created

Lex Fridman (49:06.980)

and they started to try to find projects,

Travis Oliphant (49:10.060)

people would pay for pieces

Lex Fridman (49:11.080)

and they were able to fund some of it.

Travis Oliphant (49:13.740)

Not nearly enough to keep up with what was necessary.

Lex Fridman (49:15.780)

And no criticism, just simply the reality.

Travis Oliphant (49:18.860)

I mean, it's hard to start a business

Lex Fridman (49:21.180)

and then do consulting and then also

Travis Oliphant (49:23.220)

promote an open source project that's still fairly new.

Lex Fridman (49:26.180)

Cypo is fairly niche.

Travis Oliphant (49:27.780)

We stayed connected all while I was a student,

Lex Fridman (49:30.140)

sorry, a professor.

Travis Oliphant (49:30.980)

I went to BYU and started to teach.

Lex Fridman (49:32.340)

Electrical engineering, all the applied math courses.

Travis Oliphant (49:35.060)

I loved teaching single processing,

Lex Fridman (49:36.980)

probability theory, electromagnetism.

Travis Oliphant (49:39.180)

I was, if you look at writing my professor,

Lex Fridman (49:40.940)

which my kids loved to do,

Travis Oliphant (49:42.500)

I wasn't, I got some bad reviews because people.

Lex Fridman (49:46.900)

What was the criticism?

Travis Oliphant (49:48.580)

I would speak too high of a level.

Lex Fridman (49:50.920)

Like I definitely had a calibration problem

Travis Oliphant (49:52.640)

coming out of graduate work

Lex Fridman (49:54.980)

where I hate to be condescending to people.

Travis Oliphant (49:56.980)

Like I really have a ton of respect for people fundamentally.

Lex Fridman (49:59.300)

Like my fundamental thing is I respect people.

Travis Oliphant (50:02.060)

Sometimes that can lead to a,

Lex Fridman (50:03.900)

I was thinking they had more knowledge than they did.

Lex Fridman (50:07.640)

And so I would just speak at a very high level,

Lex Fridman (50:10.100)

assume they got it.

Lex Fridman (50:11.060)

But they need to rise to the standard that you set.

Lex Fridman (50:14.340)

I mean, that's one of the,

Travis Oliphant (50:15.260)

some of the greatest teachers do that.

Lex Fridman (50:17.180)

And I agree.

Lex Fridman (50:18.020)

And that was kind of what was inspiring me.

Lex Fridman (50:19.760)

But you also have to,

Travis Oliphant (50:22.160)

I cannot say I was articulate

Lex Fridman (50:24.820)

with some of the greatest teachers, right?

Travis Oliphant (50:26.300)

I was, like one classic example,

Lex Fridman (50:28.540)

when I first taught at BYU,

Travis Oliphant (50:30.420)

my very first class, it was overheads,

Lex Fridman (50:31.980)

transparencies, overheads.

Travis Oliphant (50:34.100)

Before projectors were really that common,

Lex Fridman (50:35.940)

I taught transparencies.

Travis Oliphant (50:37.100)

I'm writing my notes out.

Lex Fridman (50:38.260)

I go in, room's half dark.

Travis Oliphant (50:40.540)

I just blaring through these transparencies.

Lex Fridman (50:42.780)

Here it is, here it is, here it is.

Lex Fridman (50:44.900)

And I did give a quiz after two weeks.

Lex Fridman (50:47.480)

No one knew anything.

Travis Oliphant (50:48.900)

Nothing I had taught had gotten anywhere.

Lex Fridman (50:50.940)

And I realized, okay, I'm not, this is not working.

Lex Fridman (50:54.140)

So I put away the transparencies

Lex Fridman (50:56.380)

and I turned around and just started using the chalkboard.

Lex Fridman (50:58.860)

And what it did is it slowed me down, right?

Lex Fridman (51:00.980)

The chalkboard just slowed me down

Lex Fridman (51:02.260)

and gave people time to process and to think.

Lex Fridman (51:04.440)

And then that made me focus.

Travis Oliphant (51:06.080)

My writing wasn't great on the chalkboard,

Lex Fridman (51:07.900)

but I really love that part of like the teaching.

Lex Fridman (51:10.520)

So that entered SciPy's world in terms of,

Lex Fridman (51:12.500)

we always understood that there's a didactic aspect

Travis Oliphant (51:14.860)

of SciPy, kind of how do you take the knowledge

Lex Fridman (51:17.740)

and then produce it?

Travis Oliphant (51:18.640)

The challenge we had was the scope.

Lex Fridman (51:21.020)

Like ultimately SciPy was everything, right?

Lex Fridman (51:23.420)

And so 2001, when it first came out,

Lex Fridman (51:25.600)

people were starting to use it.

Travis Oliphant (51:26.800)

No, this is cool, this is a tool we actually use.

Lex Fridman (51:29.580)

At the same time, 2001 timeframe,

Travis Oliphant (51:31.400)

there was a little bit of like the Hubble Space Telescope,

Lex Fridman (51:33.940)

the folks at Hubble that started to say,

Travis Oliphant (51:35.400)

hey, Python, we're gonna use Python

Lex Fridman (51:36.620)

for processing images from Hubble.

Lex Fridman (51:38.720)

And so Perry Greenfield was a good friend

Lex Fridman (51:40.820)

in running that program.

Lex Fridman (51:42.420)

And he had called me before I left WIU and said,

Lex Fridman (51:45.060)

you know, we wanna do this,

Lex Fridman (51:47.020)

but numeric actually has some challenges in terms of,

Lex Fridman (51:50.020)

you know, it's not, the array doesn't have enough types.

Travis Oliphant (51:52.700)

We need more operations.

Lex Fridman (51:54.280)

You know, broadcasting needs to be a little more settled.

Travis Oliphant (51:56.660)

They wanted record arrays.

Lex Fridman (51:57.960)

They wanted, you know, record arrays are like a data frame,

Lex Fridman (52:00.600)

but a little bit different,

Lex Fridman (52:02.220)

but they wanted more structured data.

Lex Fridman (52:03.820)

So he had called me even early on then,

Lex Fridman (52:06.020)

and he said, you know, what,

Lex Fridman (52:06.860)

would you wanna work on something to make this work?

Lex Fridman (52:08.300)

And I said, yeah, I'm interested, but I'm going here,

Lex Fridman (52:10.140)

and I, you know, we'll see if I have time.

Lex Fridman (52:12.100)

So in the meantime, while I was teaching

Lex Fridman (52:13.340)

and SciPy was emerging, and I had a student,

Lex Fridman (52:15.660)

I was constantly, while I was teaching,

Travis Oliphant (52:16.840)

trying to figure a way to fund this stuff.

Lex Fridman (52:18.840)

So I had a graduate student, my only graduate student,

Travis Oliphant (52:21.660)

a Chinese fellow, Liu Hongze is his name, great guy.

Lex Fridman (52:26.260)

He wrote a bunch of stuff for iterative linear algebra,

Travis Oliphant (52:29.900)

like got into writing some of the iterative

Lex Fridman (52:31.380)

linear algebra tools that are currently there in SciPy,

Lex Fridman (52:34.340)

and they've gotten better since,

Lex Fridman (52:36.040)

but this is in 2005, kept working on SciPy,

Lex Fridman (52:39.260)

but Perry has started working on a replacement

Lex Fridman (52:43.060)

to numeric called NumArray.

Lex Fridman (52:45.300)

And in 2004, a package called ND Image,

Lex Fridman (52:49.020)

it was an image processing library

Travis Oliphant (52:50.740)

that was written for NumArray,

Lex Fridman (52:53.220)

and it had in it a morphology tool.

Travis Oliphant (52:55.580)

I don't know if you know what morphology is.

Lex Fridman (52:56.740)

It's open, dilations, closed, you know,

Travis Oliphant (52:58.540)

there was sort of this, as a medical imaging student,

Lex Fridman (53:01.660)

I knew what it was,

Travis Oliphant (53:02.500)

because it was used in segmentation a lot.

Lex Fridman (53:04.420)

And in fact, I'd wanted to do something like that

Travis Oliphant (53:06.460)

in Python, in SciPy, but just had never gotten around to it.

Lex Fridman (53:10.220)

So when it came out, but it worked only on NumArray,

Lex Fridman (53:14.180)

and SciPy needed numeric,

Lex Fridman (53:16.420)

and so we effectively had the beginning of this split.

Lex Fridman (53:20.040)

And numeric and NumArray didn't share data,

Lex Fridman (53:22.500)

they were just two, so you could have a gigabyte

Travis Oliphant (53:24.420)

of numeric, NumArray data, and gigabyte of numeric data,

Lex Fridman (53:26.540)

and they wouldn't share it.

Lex Fridman (53:27.380)

And so you had these,

Lex Fridman (53:28.500)

then you had these scientific libraries written on top.

Travis Oliphant (53:31.300)

I got really bugged by that.

Lex Fridman (53:32.940)

I got really like, oh man, this is not good,

Travis Oliphant (53:35.060)

we're not cooperating now,

Lex Fridman (53:36.300)

we're sort of redoing each other's work,

Lex Fridman (53:37.980)

and we're just this young community.

Lex Fridman (53:40.380)

So that's what led me, even though I knew it was risky,

Travis Oliphant (53:43.940)

because my, you know, I was on a tenure track position,

Lex Fridman (53:47.140)

2004 I got reviewed.

Travis Oliphant (53:48.540)

They said, hey, things are going okay,

Lex Fridman (53:49.540)

you're doing well, paper's coming out,

Lex Fridman (53:51.540)

but you're kind of spending a lot of time

Lex Fridman (53:52.460)

doing this open source stuff, maybe do a little less of that,

Lex Fridman (53:54.780)

and a little more of the paper writing and grant writing,

Lex Fridman (53:57.260)

which was naive, but it was definitely the thinking.

Travis Oliphant (54:00.860)

It still goes on.

Lex Fridman (54:01.700)

Still goes on.

Travis Oliphant (54:03.060)

You're basically creating a thing

Lex Fridman (54:05.120)

which enables science in the 21st century.

Travis Oliphant (54:08.300)

Right.

Lex Fridman (54:09.340)

Maybe don't emphasize that so much in your free year tenure.

Travis Oliphant (54:11.980)

Right.

Lex Fridman (54:13.460)

It illustrates some of the challenges.

Travis Oliphant (54:14.860)

Yes.

Lex Fridman (54:15.700)

It does, and it's, people mean well.

Travis Oliphant (54:18.220)

Yes.

Lex Fridman (54:19.060)

Like, but we've gotten broken in a bunch of ways.

Travis Oliphant (54:22.340)

Certain things, programming,

Lex Fridman (54:23.660)

understanding the role of software engineering,

Travis Oliphant (54:25.500)

programming in society is a little bit lacking.

Lex Fridman (54:27.860)

Exactly.

Travis Oliphant (54:28.700)

Now, I was in electrical engineering position.

Lex Fridman (54:30.020)

Right.

Travis Oliphant (54:30.860)

That's even worse there.

Lex Fridman (54:33.140)

Yeah, it was very, they were very focused,

Lex Fridman (54:34.700)

and so, you know, good people, and I had a great time,

Lex Fridman (54:37.300)

I loved my time, I loved my teaching,

Travis Oliphant (54:38.940)

I loved all the things I did there.

Lex Fridman (54:40.460)

The problem was, the split was happening

Lex Fridman (54:42.540)

in this community that I loved, right?

Lex Fridman (54:43.940)

I saw people, and I went, oh my gosh,

Travis Oliphant (54:45.460)

this is gonna be, this is not great,

Lex Fridman (54:47.780)

and so I happened, you know, fate,

Travis Oliphant (54:50.020)

I had a class I had signed up for,

Lex Fridman (54:52.620)

it's a, I was trying to build an MRI system,

Lex Fridman (54:54.860)

so I had a kind of a radio, instead of a radio,

Lex Fridman (54:58.300)

a digital radio class, it was a digital MRI class.

Lex Fridman (55:01.820)

And I had people sign up, two people signed up,

Lex Fridman (55:04.020)

then they dropped, and so I had nobody in this class.

Travis Oliphant (55:06.660)

So, and I didn't have any other courses to teach,

Lex Fridman (55:08.820)

and I thought, oh, I've got some time,

Lex Fridman (55:10.940)

and I'll just write, I'll just write a replace,

Lex Fridman (55:13.100)

a merger of Numerica Numeray.

Travis Oliphant (55:14.820)

Like, I'll basically take the numeric code base

Lex Fridman (55:16.980)

at the features Numeray was adding,

Lex Fridman (55:19.220)

and then kind of come up with a single array library

Lex Fridman (55:21.180)

that everybody can use.

Lex Fridman (55:22.460)

So that's where NumPy came from,

Lex Fridman (55:24.140)

was my thinking, hey, I can do this,

Lex Fridman (55:26.500)

and who else is going to?

Lex Fridman (55:27.860)

Because at that point, I'd been around the community

Travis Oliphant (55:29.260)

long enough, and I'd written enough C code,

Lex Fridman (55:30.820)

I knew, I knew the structures, and I,

Travis Oliphant (55:33.260)

in fact, my first contribution to numeric

Lex Fridman (55:35.060)

had been writing the CAPI documentation

Travis Oliphant (55:38.580)

that went in the first documentation for NumPy,

Lex Fridman (55:41.080)

for numeric, sorry, this is Paul DuBois,

Travis Oliphant (55:43.020)

David Asher, Conrad Hinson, and myself.

Lex Fridman (55:45.100)

I got credit because I wrote this chapter,

Travis Oliphant (55:47.580)

which is all the CAPI of Numerica, all the C stuff.

Lex Fridman (55:51.260)

So I said, I'm probably the one to do it,

Lex Fridman (55:53.380)

and nobody else is gonna do this.

Lex Fridman (55:54.760)

So it was sort of, out of a sense of duty and passion,

Travis Oliphant (55:58.340)

knowing that, eh, I don't think my academic,

Lex Fridman (56:01.460)

I don't think the department here is gonna appreciate this,

Lex Fridman (56:03.940)

but it's the right thing to do.

Lex Fridman (56:06.020)

It was like.

Lex Fridman (56:06.860)

Can we just link on that moment?

Lex Fridman (56:08.660)

Yeah, yeah.

Travis Oliphant (56:09.500)

Because the importance of the way you thought

Lex Fridman (56:11.740)

and the action you took, I feel is understated

Lex Fridman (56:16.360)

and is rare and I would love to see so much more of it

Lex Fridman (56:19.900)

because what happens as the tools become more popular,

Travis Oliphant (56:24.820)

there's a split that happens.

Lex Fridman (56:27.180)

And it's a truly heroic and impactful action

Travis Oliphant (56:30.940)

to in those early, in that early split,

Lex Fridman (56:33.580)

to step up and it's like great leaders throughout history,

Travis Oliphant (56:37.820)

like get, what is the brave heart,

Lex Fridman (56:39.660)

like get on a horse and rile the troops

Travis Oliphant (56:42.500)

because I think that can have, make a big difference.

Lex Fridman (56:46.060)

We have TensorFlow versus PyTorch

Travis Oliphant (56:48.180)

in the machine learning community.

Lex Fridman (56:49.100)

We have the same problem today.

Travis Oliphant (56:50.380)

Yeah, I wonder.

Lex Fridman (56:51.780)

It's actually bigger.

Travis Oliphant (56:52.620)

I wonder if it's possible in the early days

Lex Fridman (56:56.620)

to rally the troops.

Travis Oliphant (56:58.220)

It is possible, especially in the early days.

Lex Fridman (57:00.020)

The longer it goes, the harder, right?

Travis Oliphant (57:01.620)

The more energy in the factions, the harder.

Lex Fridman (57:03.940)

But in the early days, it is possible

Lex Fridman (57:05.700)

and it's extremely helpful

Lex Fridman (57:07.660)

and there's a willingness there,

Lex Fridman (57:09.100)

but the challenge is there's just not a willingness

Lex Fridman (57:11.740)

to fund it.

Travis Oliphant (57:12.980)

There's not a willingness to, you know,

Lex Fridman (57:14.880)

like I was literally walking into a field

Travis Oliphant (57:17.540)

saying I'm going to do this

Lex Fridman (57:18.620)

and here I am, like, you know,

Travis Oliphant (57:20.140)

I have five kids at home now.

Lex Fridman (57:23.740)

Pressure builds.

Travis Oliphant (57:24.820)

Sometimes my wife hears these stories

Lex Fridman (57:26.220)

and she's like, you did what?

Travis Oliphant (57:29.020)

I thought we were going to,

Lex Fridman (57:29.860)

I thought you were actually on a path

Travis Oliphant (57:31.460)

to make sure we had resources and money, but,

Lex Fridman (57:34.100)

but again, there's a, there's an aspect,

Travis Oliphant (57:36.420)

I'm a very hopeful person.

Lex Fridman (57:37.860)

I'm an optimistic person by nature.

Travis Oliphant (57:39.680)

I love people.

Lex Fridman (57:41.120)

I learned that about myself later on.

Lex Fridman (57:43.140)

And part of my, my religious beliefs

Lex Fridman (57:47.220)

actually lead to that.

Lex Fridman (57:48.380)

And it's why I hold them dear

Lex Fridman (57:49.880)

because it's actually how I feel about,

Travis Oliphant (57:51.300)

that's what leads me to these attitudes,

Lex Fridman (57:53.420)

sort of this hopefulness and this sense of,

Travis Oliphant (57:55.900)

yeah, it may not work out for me financially

Lex Fridman (57:58.600)

or maybe, but that's not the ultimate gain.

Travis Oliphant (58:00.600)

Like that's a thing, but it's not,

Lex Fridman (58:02.940)

that's not the scorecard for me.

Lex Fridman (58:05.540)

And so I just wanted to be helpful

Lex Fridman (58:07.060)

and I knew, and partly because these SciPy conferences,

Travis Oliphant (58:09.280)

because the maintenance conversations,

Lex Fridman (58:10.860)

I knew there was a lot of need for this, right?

Lex Fridman (58:13.300)

And so I had this, it wasn't like I was alone

Lex Fridman (58:15.460)

in terms of no feedback.

Travis Oliphant (58:16.460)

I had these people who knew, but it was crazy.

Lex Fridman (58:19.440)

Like people who at the time said,

Travis Oliphant (58:20.700)

yeah, we didn't think you'd be able to do it.

Lex Fridman (58:22.340)

We thought it was crazy.

Lex Fridman (58:23.160)

And also instructive, like practically speaking,

Lex Fridman (58:26.720)

that you had a cool feature

Travis Oliphant (58:28.700)

that you were chasing the morphology, like the.

Lex Fridman (58:30.820)

Yes.

Travis Oliphant (58:31.660)

Like it's not just like.

Lex Fridman (58:32.500)

There's an end result.

Travis Oliphant (58:33.460)

It's not some visionary thing.

Lex Fridman (58:35.140)

I'm going to unite the community.

Travis Oliphant (58:36.820)

You were like. Correct.

Lex Fridman (58:38.060)

You were actually practically,

Travis Oliphant (58:39.520)

this is what one person actually could do

Lex Fridman (58:42.100)

and actually build.

Travis Oliphant (58:43.220)

Cause that is important.

Lex Fridman (58:44.220)

Cause you can get over your skis.

Travis Oliphant (58:47.460)

You can definitely get over your skis.

Lex Fridman (58:49.060)

And I had, in fact, this almost got me over my skis, right?

Travis Oliphant (58:52.140)

I would say, well, in retrospect, I hate looking back.

Lex Fridman (58:56.140)

I can tell you all the flaws with NumPy, right?

Travis Oliphant (58:58.540)

When I go into it, there's lots of stuff that I'm like,

Lex Fridman (59:00.700)

oh man, that's embarrassing.

Travis Oliphant (59:01.660)

That was wrong.

Lex Fridman (59:02.500)

I wish I had somebody stop me with a wet fish there.

Travis Oliphant (59:04.300)

Like I needed, like what I'd wished I'd had

Lex Fridman (59:07.020)

was somebody with more experience and certainly library

Travis Oliphant (59:10.460)

writing and array library.

Lex Fridman (59:11.540)

There's like, I wish I had me.

Travis Oliphant (59:12.780)

I could go back in time and go do this, do that.

Lex Fridman (59:14.520)

There's a more important thing.

Travis Oliphant (59:15.480)

Cause there's things we did that are still there

Lex Fridman (59:18.100)

that are problematic, that created challenges for later.

Lex Fridman (59:20.940)

And I didn't know it at the time.

Lex Fridman (59:22.460)

Didn't understand how important that was.

Lex Fridman (59:24.420)

And in many cases, didn't know what to do.

Lex Fridman (59:26.460)

Like there was pieces of the design of NumPy.

Travis Oliphant (59:29.060)

I didn't know what to do until five years ago.

Lex Fridman (59:31.340)

Now I know what they should have been, Ben.

Lex Fridman (59:32.860)

But I didn't know at the time and nobody,

Lex Fridman (59:33.960)

and I couldn't get the help.

Travis Oliphant (59:35.380)

Anyway, so I wrote it.

Lex Fridman (59:36.660)

It took about, it took four months to write

Travis Oliphant (59:38.780)

the first version, then about 14 months to make it usable.

Lex Fridman (59:43.360)

But it was, it wasn't, it was that first four months

Travis Oliphant (59:45.860)

of intense writing, coding, getting something out the door

Lex Fridman (59:49.320)

that worked that was, it was, it was definitely challenging.

Lex Fridman (59:52.380)

And then the big thing I did was create a new type object

Lex Fridman (59:54.900)

called D type.

Travis Oliphant (59:56.100)

That was probably the contribution.

Lex Fridman (59:58.780)

And then the fact that I added broad, not just broadcasting,