Gustav Soderstrom: Spotify - Lex Fridman 知识图谱

Lex Fridman (01:51.840)

I actually had it play at my wedding.

Gustav Soderstrom (01:55.360)

I sat with the organist and helped him play it on an organ,

Lex Fridman (01:58.400)

which was a pretty interesting experience.

Gustav Soderstrom (02:01.040)

That is probably my, I would say, top three movie of all time.

Lex Fridman (02:06.000)

Yeah, this is an incredible movie.

Gustav Soderstrom (02:07.600)

Yeah, and it came out during my formative years.

Lex Fridman (02:10.400)

And as I've discovered in music, you shape your music taste during those years.

Lex Fridman (02:15.920)

So it definitely affected me quite a bit.

Lex Fridman (02:17.840)

Did it affect you in any other kind of way?

Gustav Soderstrom (02:20.960)

Well, the movie itself affected me back then.

Lex Fridman (02:23.440)

It was a big part of culture.

Gustav Soderstrom (02:25.600)

I didn't really adopt any characters from the movie,

Lex Fridman (02:27.680)

but it was a great story of love, fantastic actors.

Lex Fridman (02:33.040)

And really, I didn't even know who Hans Zimmer was at the time, but fantastic music.

Lex Fridman (02:39.040)

And so that song has followed me.

Lex Fridman (02:42.160)

And the movie actually has followed me throughout my life.

Lex Fridman (02:43.920)

That was Quentin Tarantino, actually, I think, director or producer.

Lex Fridman (02:48.480)

So it's not Stairway to Heaven or Bohemian Rhapsody.

Lex Fridman (02:52.080)

Those are great.

Gustav Soderstrom (02:53.600)

They're not my personal favorites, but I've realized that people have different tastes.

Lex Fridman (02:57.760)

And that's a big part of what we do.

Gustav Soderstrom (03:00.400)

Well, for me, I would have to stick with Stairway to Heaven.

Lex Fridman (03:04.000)

So 35,000 years ago, I looked this up on Wikipedia,

Gustav Soderstrom (03:09.280)

flute like instruments started being used in caves as part of hunting rituals.

Lex Fridman (03:13.120)

And primitive cultural gatherings, things like that.

Gustav Soderstrom (03:15.760)

This is the birth of music.

Gustav Soderstrom (03:18.000)

Since then, we had a few folks, Beethoven, Elvis, Beatles, Justin Bieber, of course, Drake.

Lex Fridman (03:25.680)

So in your view, let's start like high level philosophical.

Lex Fridman (03:29.280)

What is the purpose of music on this planet of ours?

Gustav Soderstrom (03:35.200)

I think music has many different purposes.

Gustav Soderstrom (03:38.240)

I think there's certainly a big purpose, which is the same as much of entertainment,

Gustav Soderstrom (03:44.640)

which is escapism and to be able to live in some sort of other mental state for a while.

Lex Fridman (03:52.080)

But I also think you have the opposite of escaping,

Gustav Soderstrom (03:54.320)

which is to help you focus on something you are actually doing.

Lex Fridman (03:57.280)

Because I think people use music as a tool to tune the brain

Gustav Soderstrom (04:02.640)

to the activities that they are actually doing.

Lex Fridman (04:05.120)

And it's kind of like, in one sense, maybe it's the rawest signal.

Gustav Soderstrom (04:10.560)

If you think about the brain as neural networks,

Lex Fridman (04:13.040)

it's maybe the most efficient hack we can do to actually actively tune it

Gustav Soderstrom (04:16.880)

into some state that you want to be.

Lex Fridman (04:18.880)

You can do it in other ways.

Gustav Soderstrom (04:19.760)

You can tell stories to put people in a certain mood.

Lex Fridman (04:22.240)

But music is probably very effective to get you to a certain mood very fast, I think.

Gustav Soderstrom (04:27.120)

You know, there's a social component historically to music,

Lex Fridman (04:30.960)

where people listen to music together.

Gustav Soderstrom (04:32.480)

I was just thinking about this, that to me, and you mentioned machine learning,

Lex Fridman (04:36.880)

but to me personally, music is a really private thing.

Gustav Soderstrom (04:43.040)

I'm speaking for myself, I listen to music,

Lex Fridman (04:45.920)

like almost nobody knows the kind of things I have in my library,

Gustav Soderstrom (04:50.320)

except people who are really close to me and they really only know a certain percentage.

Lex Fridman (04:54.400)

There's like some weird stuff that I'm almost probably embarrassed by, right?

Lex Fridman (04:58.560)

It's called the guilty pleasures, right?

Lex Fridman (05:00.000)

Everyone has the guilty pleasures, yeah.

Gustav Soderstrom (05:02.560)

Hopefully they're not too bad, but for me, it's personal.

Lex Fridman (05:06.560)

Do you think of music as something that's social or as something that's personal?

Lex Fridman (05:12.880)

Or does it vary?

Lex Fridman (05:14.560)

So I think it's the same answer that you use it for both.

Gustav Soderstrom (05:20.720)

We've thought a lot about this during these 10 years at Spotify, obviously.

Lex Fridman (05:25.360)

In one sense, as you said, music is incredibly

Gustav Soderstrom (05:27.840)

social, you go to concerts and so forth.

Gustav Soderstrom (05:30.480)

On the other hand, it is your escape and everyone has these things that are very personal to them.

Lex Fridman (05:38.400)

So what we've found is that when it comes to, most people claim that they have a friend or two

Lex Fridman (05:47.680)

that they are heavily inspired by and that they listen to.

Lex Fridman (05:50.880)

So I actually think music is very social, but in a smaller group setting,

Lex Fridman (05:54.560)

it's an intimate form of, it's an intimate relationship.

Gustav Soderstrom (06:00.400)

It's not something that you necessarily share broadly.

Gustav Soderstrom (06:03.360)

Now, at concerts, you can argue you do, but then you've gathered a lot of people

Gustav Soderstrom (06:07.040)

that you have something in common with.

Gustav Soderstrom (06:08.880)

I think this broadcast sharing of music is something we tried on social networks and so forth.

Lex Fridman (06:16.960)

But it turns out that people aren't super interested in sharing their music.

Lex Fridman (06:23.120)

They aren't super interested in what their friends listen to.

Gustav Soderstrom (06:28.480)

They're interested in understanding if they have something in common perhaps with a friend,

Lex Fridman (06:32.800)

but not just as information.

Gustav Soderstrom (06:35.680)

Right, that's really interesting.

Lex Fridman (06:38.000)

I was just thinking of it this morning, listening to Spotify.

Lex Fridman (06:41.600)

I really have a pretty intimate relationship with Spotify, with my playlists, right?

Lex Fridman (06:48.480)

I've had them for many years now and they've grown with me together.

Gustav Soderstrom (06:53.360)

There's an intimate relationship you have with a library of music that you've developed.

Lex Fridman (06:59.520)

And we'll talk about different ways we can play with that.

Lex Fridman (07:02.480)

Can you do the impossible task and try to give a history of music listening

Lex Fridman (07:09.280)

from your perspective from before the internet and after the internet

Lex Fridman (07:14.160)

and just kind of everything leading up to streaming with Spotify and so on?

Lex Fridman (07:18.800)

I'll try.

Gustav Soderstrom (07:19.280)

It could be a 100 year podcast.

Lex Fridman (07:22.320)

I'll try to do a brief version.

Gustav Soderstrom (07:24.400)

There are some things that I think are very interesting during the history of music,

Lex Fridman (07:28.080)

which is that before recorded music, to be able to enjoy music,

Gustav Soderstrom (07:33.040)

you actually had to be where the music was produced

Lex Fridman (07:35.440)

because you couldn't record it and time shift it, right?

Gustav Soderstrom (07:38.640)

Creation and consumption had to happen at the same time, basically concerts.

Lex Fridman (07:41.520)

And so you either had to get to the nearest village to listen to music.

Lex Fridman (07:46.320)

And while that was cumbersome and it severely limited the distribution of music,

Lex Fridman (07:51.440)

it also had some different qualities,

Gustav Soderstrom (07:53.200)

which was that the creator could always interact with the audience.

Lex Fridman (07:56.640)

It was always live.

Lex Fridman (07:58.400)

And also there was no time cap on the music.

Lex Fridman (08:00.640)

So I think it's not a coincidence that these early classical works,

Gustav Soderstrom (08:04.960)

they're much longer than the three minutes.

Gustav Soderstrom (08:06.640)

The three minutes came in as a restriction of the first wax disc that could only contain

Lex Fridman (08:11.600)

a three minute song on one side, right?

Lex Fridman (08:14.080)

So actually the recorded music severely limited or put constraints.

Gustav Soderstrom (08:20.400)

I won't say limit.

Lex Fridman (08:21.040)

I mean, constraints are often good,

Lex Fridman (08:22.160)

but it put very hard constraints on the music format.

Lex Fridman (08:24.960)

So you kind of said, instead of doing this opus on many tens of minutes or something,

Gustav Soderstrom (08:31.200)

now you get three and a half minutes because then you're out of wax on this disc.

Lex Fridman (08:34.560)

But in return, you get an amazing distribution.

Lex Fridman (08:37.680)

Your reach will widen, right?

Lex Fridman (08:39.440)

Just on that point real quick.

Gustav Soderstrom (08:42.560)

Without the mass scale distribution, there's a scarcity component

Lex Fridman (08:47.920)

where you kind of look forward to it.

Gustav Soderstrom (08:51.760)

We had that, it's like the Netflix versus HBO Game of Thrones.

Lex Fridman (08:56.400)

You like wait for the event because you can't really listen to it.

Lex Fridman (09:00.160)

So you like look forward to it and then it's like,

Gustav Soderstrom (09:02.800)

you derive perhaps more pleasure because it's more rare for you to listen to a particular piece.

Lex Fridman (09:07.920)

You think there's value to that scarcity?

Lex Fridman (09:10.480)

Yeah, I think that that is definitely a thing.

Lex Fridman (09:12.720)

And there's always this component of if you have something in infinite amounts,

Lex Fridman (09:17.200)

will you value it as much?

Gustav Soderstrom (09:20.000)

Probably not.

Lex Fridman (09:20.880)

Humanity is always seeking some, it's relative.

Lex Fridman (09:24.400)

So you're always seeking something you didn't have.

Lex Fridman (09:25.840)

And when you have it, you don't appreciate it as much.

Lex Fridman (09:27.600)

So I think that's probably true.

Lex Fridman (09:29.520)

But I think that that's probably true.

Lex Fridman (09:31.200)

But I think that's why concerts exist.

Lex Fridman (09:33.040)

So you can actually have both.

Lex Fridman (09:35.520)

But I think net, if you couldn't listen to music in your car driving, that'd be worse.

Gustav Soderstrom (09:42.000)

That cost will be bigger than the benefit of the anticipation I think that you would have.

Gustav Soderstrom (09:47.360)

So, yeah, it started with live concerts.

Lex Fridman (09:50.720)

Then it's being able to, you know, the phonograph invented, right?

Gustav Soderstrom (09:56.720)

That you start to be able to record music.

Lex Fridman (09:59.440)

Exactly.

Lex Fridman (09:59.840)

So then you got this massive distribution that made it possible to create two things.

Gustav Soderstrom (10:04.560)

I think, first of all, cultural phenomenons, they probably need distribution to be able to happen.

Lex Fridman (10:10.560)

But it also opened access to, you know, for a new kind of artist.

Lex Fridman (10:15.520)

So you started to have these phenomenons like Beatles and Elvis and so forth.

Gustav Soderstrom (10:18.720)

That would really, a function of distribution, I think, obviously of talent and innovation.

Lex Fridman (10:23.680)

But there was also technical component.

Lex Fridman (10:25.760)

And of course, the next big innovation to come along was radio.

Lex Fridman (10:29.040)

Broadcast radio.

Lex Fridman (10:30.720)

And I think radio is interesting because it started not as a music medium.

Lex Fridman (10:36.240)

It started as an information medium for news.

Lex Fridman (10:39.600)

And then radio needed to find something to fill the time with so that they could honestly

Lex Fridman (10:45.280)

play more ads and make more money.

Lex Fridman (10:47.200)

And music was free.

Lex Fridman (10:48.480)

So then you had this massive distribution where you could program to people.

Gustav Soderstrom (10:52.480)

I think those things, that ecosystem, is what created the ability for hits.

Lex Fridman (10:59.200)

But it was also a very broadcast medium.

Lex Fridman (11:01.600)

So you would tend to get these massive, massive hits, but maybe not such a long tail.

Lex Fridman (11:07.440)

In terms of choice of everybody listens to the same stuff.

Gustav Soderstrom (11:10.480)

Yeah.

Lex Fridman (11:10.960)

And as you said, I think there are some social benefits to that.

Gustav Soderstrom (11:14.720)

I think, for example, there's a high statistical chance that if I talk about the latest episode

Lex Fridman (11:19.760)

of Game of Thrones, we have something to talk about, just statistically.

Gustav Soderstrom (11:23.280)

In the age of individual choice, maybe some of that goes away.

Lex Fridman (11:26.240)

So I do see the value of shared cultural components, but I also obviously love personalization.

Lex Fridman (11:36.400)

And so let's catch this up to the internet.

Lex Fridman (11:39.120)

So maybe Napster, well, first of all, there's MP3s, tapes, CDs.

Gustav Soderstrom (11:44.640)

There was a digitalization of music with a CD, really.

Lex Fridman (11:47.440)

It was physical distribution, but the music became digital.

Lex Fridman (11:51.200)

And so they were files, but basically boxed software, to use a software analogy.

Lex Fridman (11:56.800)

And then you could start downloading these files.

Lex Fridman (11:59.920)

And I think there are two interesting things that happened.

Gustav Soderstrom (12:02.480)

Back to music used to be longer before it was constrained by the distribution medium.

Gustav Soderstrom (12:08.080)

I don't think that was a coincidence.

Lex Fridman (12:09.840)

And then really the only music genre to have developed mostly after music was a file again

Gustav Soderstrom (12:15.600)

on the internet is EDM.

Lex Fridman (12:17.360)

And EDM is often much longer than the traditional music.

Gustav Soderstrom (12:20.640)

I think it's interesting to think about the fact that music is no longer constrained in

Lex Fridman (12:26.000)

minutes per song or something.

Gustav Soderstrom (12:27.040)

It's a legacy of an old distribution technology.

Lex Fridman (12:31.120)

And you see some of this new music that breaks the format.

Gustav Soderstrom (12:33.680)

Not so much as I would have expected actually by now, but it still happens.

Lex Fridman (12:38.160)

So first of all, I don't really know what EDM is.

Gustav Soderstrom (12:41.120)

Electronic dance music.

Lex Fridman (12:42.320)

Yeah.

Gustav Soderstrom (12:42.880)

You could say Avicii.

Lex Fridman (12:44.160)

Avicii was one of the biggest in this genre.

Lex Fridman (12:46.800)

So the main constraint is of time.

Lex Fridman (12:49.680)

Something like a three, four, five minute song.

Lex Fridman (12:52.480)

So you could have songs that were eight minutes, 10 minutes and so forth.

Lex Fridman (12:56.320)

Because it started as a digital product that you downloaded.

Lex Fridman (13:01.040)

So you didn't have this constraint anymore.

Lex Fridman (13:03.920)

So I think it's something really interesting that I don't think has fully happened yet.

Gustav Soderstrom (13:08.480)

We're kind of jumping ahead a little bit to where we are, but I think there's tons of format

Gustav Soderstrom (13:12.880)

innovation in music that should happen now, that couldn't happen when you needed to really

Gustav Soderstrom (13:18.880)

adhere to the distribution constraints.

Lex Fridman (13:20.880)

If you didn't adhere to that, you would get no distribution.

Lex Fridman (13:24.240)

So Björk, for example, the Icelandic artist, she made a full iPad app as an album.

Lex Fridman (13:30.720)

That was very expensive.

Gustav Soderstrom (13:33.440)

Even though the app store has great distribution, she gets nowhere near the distribution versus

Lex Fridman (13:38.000)

staying within the three minute format.

Lex Fridman (13:39.760)

So I think now that music is fully digital inside these streaming services, there is

Gustav Soderstrom (13:44.720)

the opportunity to change the format again and allow creators to be much more creative

Gustav Soderstrom (13:50.080)

without limiting their distribution ability.

Lex Fridman (13:52.800)

That's interesting that you're right.

Gustav Soderstrom (13:54.960)

It's surprising that we don't see that taken advantage more often.

Gustav Soderstrom (13:59.280)

It's almost like the constraints of the distribution from the 50s and 60s have molded the culture

Gustav Soderstrom (14:06.400)

to where we want the five, three to five minute song than anything else, not just.

Lex Fridman (14:12.480)

So we want the song as consumers and as artists, because I write a lot of music and I never

Gustav Soderstrom (14:18.880)

even thought about writing something longer than 10 minutes.

Lex Fridman (14:23.600)

It's really interesting that those constraints.

Lex Fridman (14:26.640)

Because all your training data has been three and a half minute songs, right?

Lex Fridman (14:29.600)

It's right.

Gustav Soderstrom (14:30.320)

Okay, so yes, digitization of data led to then mp3s.

Gustav Soderstrom (14:36.480)

Yeah, so I think you had this file then that was distributed physically, but then you had

Gustav Soderstrom (14:42.240)

the components of digital distribution and then the internet happened and there was this

Gustav Soderstrom (14:46.800)

vacuum where you had a format that could be digitally shipped, but there was no business

Gustav Soderstrom (14:51.120)

model.

Lex Fridman (14:51.840)

And then all these pirate networks happened, Napster and in Pirate Island.

Gustav Soderstrom (14:58.880)

Napster and in Sweden Pirate Bay, which was one of the biggest.

Lex Fridman (15:02.960)

And I think from a consumer point of view, which kind of leads up to the inception of

Gustav Soderstrom (15:10.080)

Spotify, from a consumer point of view, consumers for the first time had this access model to

Gustav Soderstrom (15:15.840)

music where they could, without kind of any marginal cost, they could try different tracks.

Gustav Soderstrom (15:25.680)

You could use music in new ways.

Lex Fridman (15:27.360)

There was no marginal cost.

Lex Fridman (15:28.880)

And that was a fantastic consumer experience to have access to all the music ever made,

Lex Fridman (15:32.480)

I think was fantastic.

Lex Fridman (15:34.560)

But it was also horrible for artists because there was no business model around it.

Lex Fridman (15:38.000)

So they didn't make any money.

Lex Fridman (15:39.600)

So the user need almost drove the user interface before there was a business model.

Lex Fridman (15:46.400)

And then there were these download stores that allowed you to download files, which

Gustav Soderstrom (15:52.160)

was a solution, but it didn't solve the access problem.

Lex Fridman (15:55.040)

There was still a marginal cost of 99 cents to try one more track.

Lex Fridman (15:58.560)

And I think that that heavily limits how you listen to music.

Gustav Soderstrom (16:01.920)

The example I always give is, you know, in Spotify, a huge amount of people listen to

Gustav Soderstrom (16:07.600)

music while they sleep, while they go to sleep and while they sleep.

Lex Fridman (16:11.280)

If that costed you 99 cents per three minutes, you probably wouldn't do that.

Lex Fridman (16:15.520)

And you would be much less adventurous if there was a real dollar cost to exploring

Lex Fridman (16:18.640)

music.

Lex Fridman (16:19.200)

So the access model is interesting in that it changes your music behavior.

Lex Fridman (16:22.320)

You can be, you can take much more risk because there's no marginal cost to it.

Gustav Soderstrom (16:27.680)

Maybe let me linger on piracy for a second, because I find, especially coming from Russia,

Lex Fridman (16:33.200)

piracy is something that's very interesting to me.

Gustav Soderstrom (16:39.440)

Not me, of course, ever, but I have friends who have partook in piracy of music, software,

Lex Fridman (16:49.040)

TV shows, sporting events.

Lex Fridman (16:52.400)

And usually to me, what that shows is not that they're, they can actually pay the money

Lex Fridman (16:58.400)

and they're not trying to save money.

Gustav Soderstrom (17:00.480)

They're choosing the best experience.

Lex Fridman (17:03.760)

So what to me, piracy shows is a business opportunity in all these domains.

Lex Fridman (17:08.560)

And that's where I think you're right.

Lex Fridman (17:11.120)

Spotify stepped in is basically piracy was an experience.

Gustav Soderstrom (17:15.840)

You can explore with fine music you like, and actually the interface of piracy is horrible

Gustav Soderstrom (17:23.520)

because it's, I mean, it's bad metadata, long download times, all kinds of stuff.

Lex Fridman (17:29.680)

And what Spotify does is basically first rewards artists and second makes the experience of

Lex Fridman (17:37.520)

exploring music much better.

Gustav Soderstrom (17:38.720)

I mean, the same is true, I think for movies and so on.

Gustav Soderstrom (17:42.560)

That piracy reveals in the software space, for example, I'm a huge user and fan of Adobe

Gustav Soderstrom (17:48.080)

products and there was much more incentive to pirate Adobe products before they went

Lex Fridman (17:54.720)

to a monthly subscription plan.

Lex Fridman (17:57.120)

And now all of the said friends that used to pirate Adobe products that I know now actually

Lex Fridman (18:04.640)

pay gladly for the monthly subscription.

Gustav Soderstrom (18:06.880)

Yeah, I think you're right.

Lex Fridman (18:08.000)

I think it's a sign of an opportunity for product development.

Lex Fridman (18:11.360)

And that sometimes there's a product market fit before there's a business model fit in

Lex Fridman (18:19.120)

product development.

Gustav Soderstrom (18:19.840)

I think that's a sign of it.

Lex Fridman (18:21.760)

In Sweden, I think it was a bit of both.

Gustav Soderstrom (18:24.320)

There was a culture where we even had a political party called the Pirate Party.

Lex Fridman (18:30.480)

And this was during the time when people said that information should be free.

Gustav Soderstrom (18:35.120)

It was somehow wrong to charge for ones and zeros.

Lex Fridman (18:38.080)

So I think people felt that artists should probably make some money somehow else and

Gustav Soderstrom (18:43.600)

concerts or something.

Lex Fridman (18:44.880)

So at least in Sweden, it was part really social acceptance, even at the political level.

Lex Fridman (18:49.920)

But that also forced Spotify to compete with free, which I don't think would actually

Lex Fridman (18:56.800)

could have happened anywhere else in the world.

Gustav Soderstrom (18:58.560)

The music industry needed to be doing bad enough to take that risk.

Lex Fridman (19:03.120)

And Sweden was like the perfect testing ground.

Gustav Soderstrom (19:04.800)

It had government funded high bandwidth, low latency broadband, which meant that the product

Lex Fridman (19:10.640)

would work.

Lex Fridman (19:11.440)

And it was also there was no music revenue anyway.

Lex Fridman (19:14.000)

So they were kind of like, I don't think this is going to work, but why not?

Lex Fridman (19:18.800)

So this product is one that I don't think could have happened in America, the world's

Lex Fridman (19:21.920)

largest music market, for example.

Lex Fridman (19:23.920)

So how do you compete with free?

Gustav Soderstrom (19:25.600)

Because that's an interesting world of the internet where most people don't like to

Gustav Soderstrom (19:30.640)

pay for things.

Lex Fridman (19:31.520)

So Spotify steps in and tries to, yes, compete with free.

Lex Fridman (19:36.080)

How do you do it?

Lex Fridman (19:37.120)

So I think two things.

Gustav Soderstrom (19:38.240)

One is people are starting to pay for things on the internet.

Gustav Soderstrom (19:41.680)

I think one way to think about it was that advertising was the first business model because

Gustav Soderstrom (19:47.440)

no one would put a credit card on the internet.

Lex Fridman (19:49.200)

Transactional with Amazon was the second.

Lex Fridman (19:51.600)

And maybe subscription is the third.

Lex Fridman (19:52.960)

And if you look offline, subscription is the biggest of those.

Lex Fridman (19:56.480)

So that may still happen.

Lex Fridman (19:57.600)

I think people are starting to pay for things.

Lex Fridman (19:59.040)

But definitely back then, we needed to compete with free.

Gustav Soderstrom (1:00:01.040)

today, try to, you know, make sure that their business model works, that they understand.

Gustav Soderstrom (1:00:06.080)

I think it's back to doing something to improving their products, like feedback loops and

Lex Fridman (1:00:10.880)

distribution.

Lex Fridman (1:00:11.440)

So jumping back into terms of this fascinating world of a recommender system and listening

Gustav Soderstrom (1:00:17.280)

to music and using machine learning to analyze things, do you think it's better to what

Gustav Soderstrom (1:00:24.320)

currently, correct me if I'm wrong, but currently Spotify lets people pick what they listen

Lex Fridman (1:00:30.160)

to the most part.

Gustav Soderstrom (1:00:31.680)

There's a discovery process, but you kind of organize playlists.

Gustav Soderstrom (1:00:35.040)

Is it better to let people pick what they listen to or recommend what they should listen

Lex Fridman (1:00:39.840)

to something like stations by Spotify that I saw that you're playing around with?

Lex Fridman (1:00:44.960)

Maybe you can tell me what's the status of that.

Gustav Soderstrom (1:00:47.520)

This is a Pandora style app that just kind of, as opposed to you select the music you

Lex Fridman (1:00:52.880)

listen to, it kind of feeds you the music you listen to.

Lex Fridman (1:00:58.400)

What's the status of stations by Spotify?

Lex Fridman (1:01:00.800)

What's its future?

Gustav Soderstrom (1:01:01.920)

The story of Spotify, as we have grown, has been that we made it more accessible to different

Gustav Soderstrom (1:01:07.040)

audiences and stations is another one of those where the question is, some people want to

Gustav Soderstrom (1:01:14.000)

be very specific.

Gustav Soderstrom (1:01:14.720)

They actually want to hear Starway to Heaven right now, that needs to be very easy to do.

Lex Fridman (1:01:19.760)

And some people, or even the same person, at some point might say, I want to feel upbeat

Lex Fridman (1:01:26.080)

or I want to feel happy or I want songs to sing in the car.

Lex Fridman (1:01:32.800)

So they put in the information at a very different level and then we need to translate that into

Lex Fridman (1:01:38.720)

what that means musically.

Lex Fridman (1:01:40.560)

So stations is a test to create like a consumption input vector that is much simpler where you

Lex Fridman (1:01:45.440)

can just tune it a little bit and see if that increases the overall reach.

Lex Fridman (1:01:49.520)

But we're trying to kind of serve the entire gamut of super advanced so called music aficionados

Gustav Soderstrom (1:01:56.000)

all the way to people who they love listening to music but it's not their number one priority

Gustav Soderstrom (1:02:02.560)

in life.

Lex Fridman (1:02:03.200)

They're not going to sit and follow every new release from every new artist.

Gustav Soderstrom (1:02:06.160)

They need to be able to influence music at a different level.

Lex Fridman (1:02:11.120)

So you can think of it as different products and I think one of the interesting things

Gustav Soderstrom (1:02:17.360)

to answer your question on if it's better to let the user choose or to play, I think

Gustav Soderstrom (1:02:22.080)

the answer is the challenge when machine learning kind of came along, there was a lot of thinking

Gustav Soderstrom (1:02:28.720)

about what does product development mean in a machine learning context.

Gustav Soderstrom (1:02:33.920)

People like Andrew Ng, for example, when he went to Baidu, he started doing a lot of practical

Gustav Soderstrom (1:02:38.880)

machine learning, went from academia and he thought a lot about this and he had this notion

Gustav Soderstrom (1:02:43.280)

that a product manager, designer and engineer, they used to work around this wireframe to

Gustav Soderstrom (1:02:47.760)

kind of describe what the product should look like.

Gustav Soderstrom (1:02:49.440)

It was something to talk about when you're doing a chatbot or a playlist, what are you

Lex Fridman (1:02:54.080)

going to say?

Lex Fridman (1:02:54.640)

It should be good.

Gustav Soderstrom (1:02:55.520)

That's not a good product description.

Lex Fridman (1:02:57.360)

So how do you do that?

Lex Fridman (1:02:58.400)

And he came up with this notion that the test set is the new wireframe.

Gustav Soderstrom (1:03:03.120)

The job of the product manager is to source a good test set that is representative of

Gustav Soderstrom (1:03:06.960)

what, like if you say I want to play this, that is songs to sing in the car.

Gustav Soderstrom (1:03:11.520)

The job of the product manager is to go and source a good test set of what that means.

Lex Fridman (1:03:15.360)

So then you can work with engineering to have algorithms to try to produce that.

Lex Fridman (1:03:20.000)

So we try to think a lot about how to structure product development for a machine learning

Gustav Soderstrom (1:03:25.600)

age.

Lex Fridman (1:03:26.320)

And what we discovered was that a lot of it is actually in the expectation.

Lex Fridman (1:03:30.560)

And you can go two ways.

Lex Fridman (1:03:33.120)

So let's say that if you set the expectation with the user that this is a discovery product,

Gustav Soderstrom (1:03:40.880)

like Discover Weekly, you're actually setting the expectation that most of what we show

Lex Fridman (1:03:45.280)

you will not be relevant.

Gustav Soderstrom (1:03:46.800)

When you're in the discovery process, you're going to accept that actually if you find

Lex Fridman (1:03:50.400)

one gem every Monday that you totally love, you're probably going to be happy.

Gustav Soderstrom (1:03:55.200)

Even though the statistical meaning, one out of 10 is terrible or one out of 20 is terrible

Lex Fridman (1:04:00.240)

from a user point of view because the setting was discovery is fine.

Gustav Soderstrom (1:04:03.440)

Sorry to interrupt real quick.

Gustav Soderstrom (1:04:05.360)

I just actually learned about Discover Weekly, which is a Spotify, I don't know, it's a

Gustav Soderstrom (1:04:11.600)

feature of Spotify that shows you cool songs to listen to.

Lex Fridman (1:04:16.640)

Maybe I can do issue tracking.

Gustav Soderstrom (1:04:18.160)

I couldn't find it on my Spotify app.

Lex Fridman (1:04:20.640)

It's in your library.

Gustav Soderstrom (1:04:21.680)

It's in the library.

Lex Fridman (1:04:22.640)

It's in the list of library.

Gustav Soderstrom (1:04:23.760)

Because I was like, whoa, this is cool.

Lex Fridman (1:04:25.040)

I didn't know this existed.

Lex Fridman (1:04:26.320)

And I tried to find it.

Lex Fridman (1:04:27.440)

But okay.

Gustav Soderstrom (1:04:28.800)

I will show it to you and feedback to our product team.

Lex Fridman (1:04:31.920)

There you go.

Lex Fridman (1:04:32.720)

But yeah, so yeah, sorry.

Gustav Soderstrom (1:04:34.480)

Just to mention the expectation there is basically that you're going to discover new songs.

Gustav Soderstrom (1:04:42.160)

Yeah.

Lex Fridman (1:04:42.400)

So then you can be quite adventurous in the recommendations you do.

Lex Fridman (1:04:47.920)

But we have another product called Daily Mix, which kind of implies that these are only

Lex Fridman (1:04:53.120)

going to be your favorites.

Lex Fridman (1:04:54.560)

So if you have one out of 10 that is good and nine out of 10 that doesn't work for you,

Lex Fridman (1:04:58.320)

you're going to think it's a horrible product.

Lex Fridman (1:04:59.600)

So actually a lot of the product development we learned over the years is about setting

Lex Fridman (1:05:03.040)

the right expectations.

Lex Fridman (1:05:04.080)

So for Daily Mix, you know, algorithmically, we would pick among things that feel very

Lex Fridman (1:05:09.680)

safe in your taste space.

Gustav Soderstrom (1:05:11.280)

Whereas Discover Weekly, we go kind of wild because the expectation is most of this is

Lex Fridman (1:05:15.520)

not going to.

Lex Fridman (1:05:16.400)

So a lot of that, a lot of to answer your question there, a lot of should you let the

Lex Fridman (1:05:20.960)

user pick or not?

Gustav Soderstrom (1:05:21.600)

It depends.

Gustav Soderstrom (1:05:23.360)

We have some products where the whole point is that the user can click play, put the phone

Gustav Soderstrom (1:05:26.720)

in the pocket, and it should be really good music for like an hour.

Gustav Soderstrom (1:05:30.000)

We have other products where you probably need to say like, no, no, save, no, no.

Lex Fridman (1:05:35.120)

And it's very interactive.

Lex Fridman (1:05:37.040)

I see.

Gustav Soderstrom (1:05:37.440)

That makes sense.

Lex Fridman (1:05:38.000)

And then the radio product, the stations product is one of these like click play, put in your

Gustav Soderstrom (1:05:41.920)

pocket for hours.

Lex Fridman (1:05:43.360)

That's really interesting.

Lex Fridman (1:05:44.160)

So you're thinking of different test sets for different users and trying to create products

Gustav Soderstrom (1:05:50.880)

that sort of optimize for those test sets that represent a specific set of users.

Gustav Soderstrom (1:05:57.840)

Yes, I think one thing that I think is interesting is we invested quite heavily in editorial

Lex Fridman (1:06:06.160)

in people creating playlists using statistical data.

Lex Fridman (1:06:09.520)

And that was successful for us.

Lex Fridman (1:06:10.800)

And then we also invested in machine learning.

Lex Fridman (1:06:13.600)

And for the longest time within Spotify and within the rest of the industry, there was

Lex Fridman (1:06:18.000)

always this narrative of humans versus the machine, algo versus editorial.

Lex Fridman (1:06:23.360)

And editors would say like, well, if I had that data, if I could see your

Gustav Soderstrom (1:06:27.600)

playlisting history and I made a choice for you, I would have made a better choice.

Lex Fridman (1:06:31.680)

And they would have because they're much smarter than these algorithms.

Lex Fridman (1:06:35.200)

The human is incredibly smart compared to our algorithms.

Gustav Soderstrom (1:06:38.880)

They can take culture into account and so forth.

Gustav Soderstrom (1:06:41.440)

The problem is that they can't make 200 million decisions per hour for every user that logs

Gustav Soderstrom (1:06:47.600)

in.

Lex Fridman (1:06:47.680)

So the algo may be not as sophisticated, but much more efficient.

Lex Fridman (1:06:51.760)

So there was this contradiction.

Lex Fridman (1:06:54.480)

But then a few years ago, we started focusing on this kind of human in the loop thinking

Gustav Soderstrom (1:07:00.160)

around machine learning.

Lex Fridman (1:07:01.280)

And we actually coined an internal term for it called algotorial, a combination of algorithms

Lex Fridman (1:07:07.120)

and editors, where if we take a concrete example, you think of the editor, this paid

Gustav Soderstrom (1:07:15.040)

expert that we have that's really good at something like soul, hip hop, EDM, something,

Lex Fridman (1:07:20.400)

right?

Lex Fridman (1:07:20.720)

They're a true expert, no one in the industry.

Lex Fridman (1:07:22.800)

So they have all the cultural knowledge.

Lex Fridman (1:07:24.480)

You think of them as the product manager.

Lex Fridman (1:07:26.560)

And you say that, let's say that you want to create a, you think that there's a product

Gustav Soderstrom (1:07:32.880)

need in the world for something like songs to sing in the car or songs to sing in the

Gustav Soderstrom (1:07:36.160)

shower.

Lex Fridman (1:07:36.560)

I'm taking that example because it exists.

Lex Fridman (1:07:38.400)

People love to scream songs in the car when they drive, right?

Lex Fridman (1:07:42.560)

So you want to create that product and you have this product manager who's a musical

Gustav Soderstrom (1:07:45.520)

expert.

Gustav Soderstrom (1:07:46.640)

They create, they come up with a concept, like I think this is a missing thing in humanity,

Gustav Soderstrom (1:07:50.800)

like a playlist called songs to sing in the car.

Gustav Soderstrom (1:07:53.920)

They create the framing, the image, the title, and they create a test set of, they create

Gustav Soderstrom (1:07:59.840)

a group of songs, like a few thousand songs out of the catalog that they manually curate

Lex Fridman (1:08:04.480)

that are known songs that are great to sing in the car.

Lex Fridman (1:08:07.520)

And they can take like true romance into account.

Lex Fridman (1:08:09.840)

They understand things that our algorithms do not at all.

Lex Fridman (1:08:12.400)

So they have this huge set of tracks.

Gustav Soderstrom (1:08:14.480)

Then when we deliver that to you, we look at your taste vectors and you get the 20 tracks

Gustav Soderstrom (1:08:19.600)

that are songs to sing in the car in your taste.

Lex Fridman (1:08:22.560)

So you have personalization and editorial input in the same process, if that makes sense.

Gustav Soderstrom (1:08:29.520)

Yeah, it makes total sense.

Lex Fridman (1:08:30.880)

And I have several questions around that.

Gustav Soderstrom (1:08:32.480)

This is like fascinating.

Lex Fridman (1:08:36.080)

Okay.

Lex Fridman (1:08:36.560)

So first, it is a little bit surprising to me that the world expert humans are outperforming

Lex Fridman (1:08:44.720)

machines at specifying songs to sing in the car.

Lex Fridman (1:08:50.960)

So maybe you could talk to that a little bit.

Lex Fridman (1:08:53.680)

I don't know if you can put it into words, but what is it?

Lex Fridman (1:08:57.760)

How difficult is this problem?

Lex Fridman (1:09:01.680)

Do you really, I guess what I'm trying to ask is there, how difficult is it to encode

Lex Fridman (1:09:06.720)

the cultural references, the context of the song, the artists, all those things together?

Lex Fridman (1:09:14.640)

Can machine learning really not do that?

Gustav Soderstrom (1:09:17.360)

I mean, I think machine learning is great at replicating patterns if you have the patterns.

Lex Fridman (1:09:23.040)

But if you try to write with me a spec of what song's greatest song to sing in the car

Lex Fridman (1:09:27.680)

definition is, is it loud?

Lex Fridman (1:09:30.320)

Does it have many choruses?

Lex Fridman (1:09:31.520)

Should it have been in movies?

Lex Fridman (1:09:32.800)

It quickly gets incredibly complicated, right?

Gustav Soderstrom (1:09:35.680)

Yeah.

Lex Fridman (1:09:36.880)

And a lot of it may not be in the structure of the song or the title.

Gustav Soderstrom (1:09:40.960)

It could be cultural references because, you know, it was a history.

Lex Fridman (1:09:44.880)

So the definition problems quickly get, and I think that was the insight of Andrew Ng

Gustav Soderstrom (1:09:51.360)

when he said the job of the product manager is to understand these things that algorithms

Lex Fridman (1:09:55.440)

don't and then define what that looks like.

Lex Fridman (1:09:58.640)

And then you have something to train towards, right?

Lex Fridman (1:10:00.880)

Then you have kind of the test set.

Lex Fridman (1:10:02.720)

And then so today the editors create this pool of tracks and then we personalize.

Gustav Soderstrom (1:10:06.960)

You could easily imagine that once you have this set, you could have some automatic exploration

Gustav Soderstrom (1:10:11.120)

on the rest of the catalog because then you understand what it is.

Lex Fridman (1:10:14.480)

And then the other side of it, when machine learning does help is this taste vector.

Lex Fridman (1:10:20.560)

How hard is it to construct a vector that represents the things an individual human

Lex Fridman (1:10:26.960)

likes, this human preference?

Lex Fridman (1:10:30.080)

So you can, you know, music isn't like, it's not like Amazon, like things you usually buy.

Lex Fridman (1:10:38.320)

Music seems more amorphous.

Gustav Soderstrom (1:10:39.920)

Like it's this thing that's hard to specify.

Lex Fridman (1:10:42.560)

Like what is, you know, if you look at my playlist, what is the music that I love?

Gustav Soderstrom (1:10:48.080)

It's harder.

Lex Fridman (1:10:49.360)

It seems to be much more difficult to specify concretely.

Lex Fridman (1:10:54.080)

So how hard is it to build a taste vector?

Lex Fridman (1:10:57.120)

It is very hard in the sense that you need a lot of data.

Lex Fridman (1:11:00.720)

And I think what we found was that, so it's not a stationary problem.

Lex Fridman (1:11:06.240)

It changes over time.

Lex Fridman (1:11:08.720)

And so we've gone through the journey of, if you've done a lot of computer vision,

Lex Fridman (1:11:15.680)

obviously I've done a bunch of computer vision in my past.

Lex Fridman (1:11:18.320)

And we started kind of with the handcrafted heuristics for, you know, this is kind of

Lex Fridman (1:11:24.160)

indie music.

Gustav Soderstrom (1:11:24.800)

This is this.

Lex Fridman (1:11:25.360)

And if you consume this, you'd probably like this.

Lex Fridman (1:11:27.440)

So we have, we started there and we have some of that still.

Gustav Soderstrom (1:11:31.200)

Then what was interesting about the playlist data was that you could find these latent

Gustav Soderstrom (1:11:34.720)

things that wouldn't necessarily even make sense to you.

Lex Fridman (1:11:38.800)

That could even capture maybe cultural references because they cooccurred.

Gustav Soderstrom (1:11:42.880)

Things that wouldn't have appeared kind of mechanistically either in the content or so

Lex Fridman (1:11:48.160)

forth.

Lex Fridman (1:11:48.400)

So I think that, I think the core assumption is that there are patterns in almost

Lex Fridman (1:12:01.280)

everything.

Lex Fridman (1:12:02.640)

And if there are patterns, these embedding techniques are getting better and better now.

Gustav Soderstrom (1:12:06.960)

Now, as everyone else, we're also using kind of deep embeddings where you can encode

Gustav Soderstrom (1:12:12.400)

binary values and so forth.

Lex Fridman (1:12:14.400)

And what I think is interesting is this process to try to find things that do not

Gustav Soderstrom (1:12:21.280)

necessarily, you wouldn't actually have guessed.

Lex Fridman (1:12:23.920)

So it is very hard in an engineering sense to find the right dimensions.

Gustav Soderstrom (1:12:28.560)

It's an incredible scalability problem to do for hundreds of millions of users and to

Lex Fridman (1:12:33.920)

update it every day.

Lex Fridman (1:12:35.920)

But in theory, in theory embeddings isn't that complicated.

Gustav Soderstrom (1:12:42.160)

The fact that you try to find some principal components or something like that, dimensionality

Gustav Soderstrom (1:12:46.240)

reduction and so forth.

Lex Fridman (1:12:47.040)

So the theory, I guess, is easy.

Gustav Soderstrom (1:12:48.240)

The practice is very, very hard.

Lex Fridman (1:12:50.480)

And it's a huge engineering challenge.

Lex Fridman (1:12:53.120)

But fortunately, we have some amazing both research and engineering teams in this space.

Lex Fridman (1:12:58.400)

Yeah, I guess the question is all, I mean, it's similar.

Gustav Soderstrom (1:13:03.200)

I deal with it with autonomous vehicle spaces.

Lex Fridman (1:13:05.360)

The question is how hard is driving?

Lex Fridman (1:13:07.680)

And here is basically the question is of edge cases.

Lex Fridman (1:13:14.560)

So embedding probably works, not probably, but I would imagine works well in a lot of

Gustav Soderstrom (1:13:22.240)

cases.

Lex Fridman (1:13:24.000)

So there's a bunch of questions that arise then.

Lex Fridman (1:13:25.840)

So do song preferences, does your taste vector depend on context, like mood, right?

Lex Fridman (1:13:33.760)

So there's different moods, and so how does that take in it?

Lex Fridman (1:13:41.840)

Is it possible to take that as a consideration?

Lex Fridman (1:13:44.320)

Or do you just leave that as a interface problem that allows the user to just control it?

Lex Fridman (1:13:49.840)

So when I'm looking for workout music, I kind of specify it by choosing certain playlists,

Lex Fridman (1:13:55.440)

doing certain search.

Gustav Soderstrom (1:13:56.560)

Yeah, so that's a great point.

Lex Fridman (1:13:58.560)

Back to the product development.

Gustav Soderstrom (1:14:00.080)

You could try to spend a few years trying to predict which mood you're in automatically

Lex Fridman (1:14:04.480)

when you open Spotify, or you create a tab which is happy and sad, right?

Lex Fridman (1:14:08.320)

And you're going to be right 100% of the time with one click.

Gustav Soderstrom (1:14:10.880)

Now, it's probably much better to let the user tell you if they're happy or sad, or

Gustav Soderstrom (1:14:14.880)

if they want to work out.

Gustav Soderstrom (1:14:15.840)

On the other hand, if your user interface becomes 2,000 tabs, you're introducing so

Gustav Soderstrom (1:14:20.480)

much friction so no one will use the product.

Lex Fridman (1:14:22.080)

So then you have to get better.

Lex Fridman (1:14:24.080)

So it's this thing where you have to be able to get better.

Lex Fridman (1:14:26.800)

So then you have to get better, so it's this thing where I think maybe it was, I don't

Lex Fridman (1:14:32.640)

remember who coined it, but it's called fault tolerant UIs, right?

Gustav Soderstrom (1:14:35.040)

You build a UI that is tolerant of being wrong, and then you can be much less right in your

Gustav Soderstrom (1:14:42.000)

algorithms.

Lex Fridman (1:14:43.120)

So we've had to learn a lot of that.

Gustav Soderstrom (1:14:45.440)

Building the right UI that fits where the machine learning is, and a great discovery

Gustav Soderstrom (1:14:52.160)

there, which was by the teams during one of our hack days, was this thing of taking discovery,

Gustav Soderstrom (1:14:58.720)

packaging it into a playlist, and saying that these are new tracks that we think you might

Lex Fridman (1:15:04.880)

like based on this.

Lex Fridman (1:15:05.920)

And setting the right expectation made it a great product.

Lex Fridman (1:15:09.440)

So I think we have this benefit that, for example, Tesla doesn't have that we can change

Gustav Soderstrom (1:15:15.920)

the expectation.

Lex Fridman (1:15:16.800)

We can build a fault tolerant setting.

Gustav Soderstrom (1:15:18.640)

It's very hard to be fault tolerant when you're driving at 100 miles per hour or something.

Lex Fridman (1:15:23.760)

And we have the luxury of being able to say that of being wrong if we have the right UI,

Gustav Soderstrom (1:15:30.000)

which gives us different abilities to take more risk.

Lex Fridman (1:15:33.440)

So I actually think the self driving problem is much harder.

Gustav Soderstrom (1:15:37.680)

Oh, yeah, for sure.

Lex Fridman (1:15:39.680)

It's much less fun because people die.

Gustav Soderstrom (1:15:44.240)

Exactly.

Lex Fridman (1:15:45.200)

And in Spotify, it's such a more fun problem because failure is beautiful in a way.

Gustav Soderstrom (1:15:55.040)

It leads to exploration.

Lex Fridman (1:15:56.320)

So it's a really fun reinforcement learning problem.

Lex Fridman (1:15:58.640)

The worst case scenario is you get these WTF tweets like, how did I get this?

Lex Fridman (1:16:02.800)

This song, yeah.

Gustav Soderstrom (1:16:03.600)

Which is a lot better than the self driving.

Gustav Soderstrom (1:16:05.440)

Exactly, so what's the feedback that a user, what's the signal that a user provides into

Lex Fridman (1:16:14.400)

the system?

Lex Fridman (1:16:15.440)

So you mentioned skipping.

Lex Fridman (1:16:19.360)

What is like the strongest signal?

Lex Fridman (1:16:22.000)

You didn't mention clicking like.

Lex Fridman (1:16:24.800)

So we have a few signals that are important.

Lex Fridman (1:16:27.600)

Obviously playing, playing through.

Lex Fridman (1:16:30.240)

So one of the benefits of music, actually, even compared to podcasts or movies is the

Lex Fridman (1:16:36.560)

object itself is really only about three minutes.

Lex Fridman (1:16:39.280)

So you get a lot of chances to recommend and the feedback loop is every three minutes instead

Lex Fridman (1:16:44.320)

of every two hours or something.

Lex Fridman (1:16:45.760)

So you actually get kind of noisy, but quite fast feedback.

Lex Fridman (1:16:50.880)

And so you can see if people play through, which is the inverse of skip really.

Gustav Soderstrom (1:16:55.200)

That's an important signal.

Gustav Soderstrom (1:16:56.560)

On the other hand, much of the consumption happens when your phone is in your pocket.

Gustav Soderstrom (1:17:00.320)

Maybe you're running or driving or you're playing on a speaker.

Lex Fridman (1:17:03.040)

And so you not skipping doesn't mean that you love that song.

Gustav Soderstrom (1:17:05.600)

It may be that it wasn't bad enough that you would walk up and skip.

Lex Fridman (1:17:08.960)

So it's a noisy signal.

Gustav Soderstrom (1:17:10.560)

Then we have the equivalent of the like, which is you saved it to your library.

Lex Fridman (1:17:14.000)

That's a pretty strong signal of affection.

Lex Fridman (1:17:16.720)

And then we have the more explicit signal of playlisting.

Lex Fridman (1:17:21.280)

Like you took the time to create a playlist, you put it in there.

Gustav Soderstrom (1:17:23.920)

There's a very little small chance that if you took all that trouble, this is not a really

Lex Fridman (1:17:28.960)

important track to you.

Lex Fridman (1:17:30.480)

And then we understand also what are the tracks it relates to.

Lex Fridman (1:17:34.000)

So we have the playlisting, we have the like, and then we have the listening or skip.

Lex Fridman (1:17:39.120)

And you have to have very different approaches to all of them because of different levels

Lex Fridman (1:17:43.360)

of noise.

Gustav Soderstrom (1:17:44.400)

One is very voluminous, but noisy, and the other is rare, but you can probably trust it.

Gustav Soderstrom (1:17:49.760)

Yeah, it's interesting because I think between those signals captures all the information

Gustav Soderstrom (1:17:55.680)

you'd want to capture.

Gustav Soderstrom (1:17:57.040)

I mean, there's a feeling, a shallow feeling for me that there's sometimes that I'll hear

Gustav Soderstrom (1:18:01.520)

a song that's like, yes, this is, you know, this was the right song for the moment.

Lex Fridman (1:18:05.920)

But there's really no way to express that fact except by listening through it all the

Gustav Soderstrom (1:18:10.720)

way and maybe playing it again at that time or something.

Lex Fridman (1:18:14.240)

But there's no need for a button that says this was the best song I could have heard

Gustav Soderstrom (1:18:19.680)

at this moment.

Gustav Soderstrom (1:18:20.400)

Well, we're playing around with that, with kind of the thumbs up concept saying like,

Gustav Soderstrom (1:18:24.080)

I really like this.

Lex Fridman (1:18:25.200)

Just kind of talking to the algorithm.

Gustav Soderstrom (1:18:27.520)

It's unclear if that's the best way for humans to interact.

Lex Fridman (1:18:30.640)

Maybe it is.

Gustav Soderstrom (1:18:31.200)

Maybe they should think of Spotify as a person, an agent sitting there trying to serve you

Lex Fridman (1:18:35.600)

and you can say like, bad Spotify, good Spotify.

Gustav Soderstrom (1:18:38.720)

Right now, the analogy we've had is more, you shouldn't think of us.

Lex Fridman (1:18:42.880)

We should be invisible.

Lex Fridman (1:18:44.400)

And the feedback is if you save it, it's kind of you work for yourself.

Lex Fridman (1:18:48.320)

You do a playlist because you think it's great and we can learn from that.

Gustav Soderstrom (1:18:50.960)

It's kind of back to Tesla, how they kind of have this shadow mode.

Lex Fridman (1:18:55.200)

They sit in what you drive.

Gustav Soderstrom (1:18:56.720)

We kind of took the same analogy.

Gustav Soderstrom (1:18:58.560)

We sit in what you playlist and then maybe we can offer you an autopilot where you can

Gustav Soderstrom (1:19:02.800)

take over for a while or something like that.

Lex Fridman (1:19:04.640)

And then back off if you say like, that's not good enough.

Lex Fridman (1:19:08.240)

But I think it's interesting to figure out what your mental model is.

Gustav Soderstrom (1:19:11.600)

If Spotify is an AI that you talk to, which I think might be a bit too abstract for many

Gustav Soderstrom (1:19:18.880)

consumers, or if you still think of it as it's my music app, but it's just more helpful.

Lex Fridman (1:19:24.320)

And it depends on the device it's running on, which brings us to smart speakers.

Lex Fridman (1:19:31.040)

So I have a lot of the Spotify listening I do is on devices I can talk to, whether it's

Lex Fridman (1:19:38.400)

from Amazon, Google or Apple.

Lex Fridman (1:19:39.920)

What's the role of Spotify on those devices?

Lex Fridman (1:19:42.320)

How do you think of it differently than on the phone or on the desktop?

Gustav Soderstrom (1:19:47.840)

There are a few things to say about the first of all, it's incredibly exciting.

Lex Fridman (1:19:52.080)

They're growing like crazy, especially here in the US.

Lex Fridman (1:19:58.320)

And it's solving a consumer need that I think is, you can think of it as just remote interactivity.

Lex Fridman (1:20:09.200)

You can control this thing from across the room.

Lex Fridman (1:20:11.840)

And it may feel like a small thing, but it turns out that friction matters to consumers

Gustav Soderstrom (1:20:16.880)

being able to say play, pause and so forth from across the room is very powerful.

Lex Fridman (1:20:22.000)

So basically, you made the living room interactive now.

Lex Fridman (1:20:26.000)

And what we see in our data is that the number one use case for these speakers is music,

Gustav Soderstrom (1:20:33.600)

music and podcast.

Lex Fridman (1:20:34.960)

So fortunately for us, it's been important to these companies to have those use case

Gustav Soderstrom (1:20:39.920)

covered.

Lex Fridman (1:20:40.640)

So they want to Spotify on this.

Gustav Soderstrom (1:20:42.080)

We have very good relationships with them.

Lex Fridman (1:20:45.840)

And we're seeing tremendous success with them.

Lex Fridman (1:20:51.200)

What I think is interesting about them is it's already working.

Lex Fridman (1:20:57.360)

We kind of had this epiphany many years ago, back when we started using Sonos.

Gustav Soderstrom (1:21:02.720)

If you went through all the trouble of setting up your Sonos system, you had this magical

Lex Fridman (1:21:06.800)

experience where you had all the music ever made in your living room.

Lex Fridman (1:21:10.400)

And we made this assumption that the home, everyone used to have a CD player at home,

Lex Fridman (1:21:16.320)

but they never managed to get their files working in the home.

Gustav Soderstrom (1:21:19.040)

Having this network attached storage was too cumbersome for most consumers.

Lex Fridman (1:21:22.960)

So we made the assumption that the home would skip from the CD all the way to streaming

Gustav Soderstrom (1:21:26.480)

books, where you would buy the steering and would have all the music built in.

Lex Fridman (1:21:31.120)

That took longer than we thought.

Lex Fridman (1:21:32.640)

But with the voice speakers, that was the unlocking that made kind of the connected

Lex Fridman (1:21:36.080)

speaker happen in the home.

Lex Fridman (1:21:39.760)

So it really exploded.

Lex Fridman (1:21:41.520)

And we saw this engagement that we predicted would happen.

Lex Fridman (1:21:45.760)

What I think is interesting, though, is where it's going from now.

Lex Fridman (1:21:49.120)

Right now, you think of them as voice speakers.

Lex Fridman (1:21:51.920)

But I think if you look at Google I.O., for example, they just added a camera to it, where

Gustav Soderstrom (1:21:58.640)

when the alarm goes off, instead of saying, hey, Google, stop, you can just wave your

Gustav Soderstrom (1:22:04.240)

hand.

Lex Fridman (1:22:05.040)

So I think they're going to think more of it as an agent or as an assistant, truly an

Gustav Soderstrom (1:22:11.920)

assistant.

Lex Fridman (1:22:12.400)

And an assistant that can see you is going to be much more effective than a blind assistant.

Lex Fridman (1:22:17.040)

So I think these things will morph.

Lex Fridman (1:22:18.480)

And we won't necessarily think of them as, quote unquote, voice speakers anymore.

Gustav Soderstrom (1:22:22.560)

Just as interactive access to the Internet in the home.

Lex Fridman (1:22:29.200)

But I still think that the biggest use case for those will be audio.

Lex Fridman (1:22:34.080)

So for that reason, we're investing heavily in it.

Lex Fridman (1:22:36.640)

And we built our own NLU stack to be able to the challenge here is, how do you innovate

Lex Fridman (1:22:43.520)

in that world?

Lex Fridman (1:22:44.240)

It lowers friction for consumers, but it's also much more constrained.

Gustav Soderstrom (1:22:48.320)

You have no pixels to play with in an audio only world.

Lex Fridman (1:22:51.600)

It's really the vocabulary that is the interface.

Lex Fridman (1:22:54.880)

So we started investing and playing around quite a lot with that, trying to understand

Lex Fridman (1:22:58.560)

what the future will be of you speaking and gesturing and waving at your music.

Lex Fridman (1:23:03.360)

And actually, you're actually nudging closer to the autonomous vehicle space because from

Gustav Soderstrom (1:23:08.480)

everything I've seen, the level of frustration people experience upon failure of natural

Gustav Soderstrom (1:23:14.080)

language understanding is much higher than failure in other contexts.

Lex Fridman (1:23:18.320)

People get frustrated really fast.

Lex Fridman (1:23:20.400)

So if you screw that experience up even just a little bit, they give up really quickly.

Lex Fridman (1:23:25.600)

Yeah.

Lex Fridman (1:23:26.320)

And I think you see that in the data.

Gustav Soderstrom (1:23:28.320)

While it's tremendously successful, the most common interactions are play, pause and next.

Gustav Soderstrom (1:23:36.160)

The things where if you compare it to taking up your phone, unlocking it, bringing up the

Lex Fridman (1:23:39.440)

app and skipping, clicking skip, it was much lower friction.

Lex Fridman (1:23:44.160)

But then for longer, more complicated things like, can you find me that song about the

Lex Fridman (1:23:49.280)

people still bring up the phone and search and then play it on their speaker?

Lex Fridman (1:23:51.920)

So we tried again to build a fault tolerant UI where for the more complicated things,

Gustav Soderstrom (1:23:56.960)

you can still pick up your phone, have powerful full keyboard search and then try to optimize

Gustav Soderstrom (1:24:02.480)

for where there is actually lower friction and try to it's kind of like the test autopilot

Lex Fridman (1:24:07.280)

thing.

Gustav Soderstrom (1:24:07.840)

You have to be at the level where you're helpful.

Lex Fridman (1:24:11.040)

If you're too smart and just in the way, people are going to get frustrated.

Lex Fridman (1:24:15.040)

And first of all, I'm not obsessed with stairway to heaven.

Lex Fridman (1:24:18.080)

It's just a good song.

Lex Fridman (1:24:19.440)

But let me mention that as a use case because it's an interesting one.

Gustav Soderstrom (1:24:22.880)

I've literally told one of I don't want to say the name of the speaker because when people

Gustav Soderstrom (1:24:28.160)

are listening to it, it'll make their speaker go off.

Lex Fridman (1:24:30.320)

But I talked to the speaker and I say play stairway to heaven.

Lex Fridman (1:24:34.720)

And every time it like not every time, but a large percentage of the time plays the wrong

Lex Fridman (1:24:40.320)

stairway to heaven.

Gustav Soderstrom (1:24:41.440)

It plays like some cover of the and that part of the experience.

Gustav Soderstrom (1:24:48.240)

I actually wonder from a business perspective, does Spotify control that entire experience

Lex Fridman (1:24:55.120)

or no?

Gustav Soderstrom (1:24:56.160)

It seems like the NLU, the natural language stuff is controlled by the speaker and then

Gustav Soderstrom (1:25:01.680)

Spotify stays at a layer below that.

Lex Fridman (1:25:04.640)

It's a good and complicated question.

Gustav Soderstrom (1:25:07.040)

Some of which is dependent on the on the partners.

Lex Fridman (1:25:11.200)

So it's hard to comment on the on the specifics.

Lex Fridman (1:25:13.280)

But the question is the right one.

Gustav Soderstrom (1:25:15.840)

The challenge is if you can't use any of the personalization, I mean, we know which stairway

Gustav Soderstrom (1:25:21.280)

to heaven.

Lex Fridman (1:25:21.840)

And the truth is maybe for for one person, it is exactly the cover that they want.

Lex Fridman (1:25:26.400)

And they would be very frustrated if a place I think we I think we default to the right

Lex Fridman (1:25:31.440)

version.

Lex Fridman (1:25:31.760)

But but you actually want to be able to do the cover for the person that just played

Lex Fridman (1:25:35.280)

the cover 50 times.

Gustav Soderstrom (1:25:36.320)

Or Spotify is just going to seem stupid.

Lex Fridman (1:25:38.400)

So you want to be able to leverage the personalization.

Lex Fridman (1:25:40.160)

But you have this stack where you have the the ASR and this thing called the end best

Lex Fridman (1:25:46.320)

list of the best guesses here.

Lex Fridman (1:25:48.480)

And then the position comes in at the end.

Gustav Soderstrom (1:25:50.480)

You actually want the person to be here when you're guessing about what they actually

Gustav Soderstrom (1:25:53.280)

meant.

Lex Fridman (1:25:54.000)

So we're working with these partners and it's a complicated it's a complicated thing

Gustav Soderstrom (1:26:00.160)

where you want to you want to be able.

Lex Fridman (1:26:02.880)

So first of all, you want to be very careful with your users data.

Gustav Soderstrom (1:26:06.800)

You don't want to share your users data without the permission.

Lex Fridman (1:26:09.200)

But you want to share some data so that their experience gets better.

Lex Fridman (1:26:12.640)

So that these partners can understand enough, but not too much and so forth.

Lex Fridman (1:26:16.400)

So it's really the trick is that it's like a business driven relationship where you're

Gustav Soderstrom (1:26:21.760)

doing product development across companies together, which is which is really complicated.

Lex Fridman (1:26:26.960)

But this is exactly why we built our own NLU so that we actually can make personalized

Gustav Soderstrom (1:26:32.960)

guesses, because this is the biggest frustration from a user point of view.

Lex Fridman (1:26:36.320)

They don't understand about ASR and best list and and business deals.

Lex Fridman (1:26:40.160)

They're like, how hard can it be?

Gustav Soderstrom (1:26:41.440)

I was told this thing 50 times this version and still the place the wrong thing.

Gustav Soderstrom (1:26:45.120)

It can't it can't be hard.

Lex Fridman (1:26:47.040)

So we try to take the user approach.

Gustav Soderstrom (1:26:48.640)

If the user the user is not going to understand the complications of business, we have to

Lex Fridman (1:26:53.360)

solve it.

Lex Fridman (1:26:53.760)

So let's talk about sort of a complicated subject that I myself I'm quite torn about

Lex Fridman (1:27:02.960)

the idea sort of of paying artists.

Gustav Soderstrom (1:27:08.640)

Right.

Gustav Soderstrom (1:27:09.840)

I saw as of August 31st, 2018, over 11 billion dollars were paid to rights holders.

Lex Fridman (1:27:17.200)

So and further distributed to artists from Spotify.

Lex Fridman (1:27:21.200)

So a lot of money is being paid to artists.

Gustav Soderstrom (1:27:23.840)

First of all, the whole time as a consumer for me, when I look at Spotify, I'm not sure

Gustav Soderstrom (1:27:30.800)

I'm remembering correctly, but I think you said exactly how I feel, which is this is

Gustav Soderstrom (1:27:34.880)

too good to be true.

Gustav Soderstrom (1:27:36.240)

Like when I start using Spotify, I assume you guys will go bankrupt in like a month.

Gustav Soderstrom (1:27:43.040)

It's like this is too good.

Lex Fridman (1:27:44.400)

A lot of people did.

Gustav Soderstrom (1:27:47.040)

I was like, this is amazing.

Lex Fridman (1:27:48.960)

So one question I have is sort of the bigger question.

Lex Fridman (1:27:53.200)

How do you make money in this complicated world?

Lex Fridman (1:27:55.840)

How do you deal with the relationship with record labels who are complicated?

Gustav Soderstrom (1:28:04.800)

These big you're essentially have the task of herding cats, but like rich and powerful

Gustav Soderstrom (1:28:14.080)

powerful cats, and also have the task of paying artists enough and paying those labels enough

Lex Fridman (1:28:21.520)

and still making money in the Internet space where people are not willing to pay hundreds

Lex Fridman (1:28:26.480)

of dollars a month.

Lex Fridman (1:28:27.920)

So how do you navigate the space?

Lex Fridman (1:28:30.720)

How do you navigate?

Gustav Soderstrom (1:28:31.600)

That's a beautiful description.

Lex Fridman (1:28:32.560)

Herding rich cats.

Gustav Soderstrom (1:28:34.720)

That before.

Gustav Soderstrom (1:28:37.200)

It is very complicated, and I think certainly actually betting against Spotify has been

Gustav Soderstrom (1:28:42.880)

statistically a very smart thing to do.

Gustav Soderstrom (1:28:45.040)

Just looking at the at the line of roadkill in music streaming services, it's it's kind

Gustav Soderstrom (1:28:52.880)

of I think if I understood the complexity when I joined Spotify, unfortunately, fortunately,

Gustav Soderstrom (1:28:58.560)

I didn't know enough about the music industry to understand the complexities, because then

Gustav Soderstrom (1:29:03.440)

I would have made a more rational guess that it wouldn't work.

Lex Fridman (1:29:06.240)

So, you know, ignorance is bliss.

Lex Fridman (1:29:08.480)

But I think there have been a few distinct challenges.

Gustav Soderstrom (1:29:13.200)

I think, as I said, one of the things that made it work at all was that Sweden and the

Gustav Soderstrom (1:29:17.600)

Nordics was a lost market.

Lex Fridman (1:29:19.840)

So there was no risk for labels to try this.

Gustav Soderstrom (1:29:25.120)

I don't think it would have worked if if the market was healthy.

Lex Fridman (1:29:29.760)

So that was the initial condition.

Gustav Soderstrom (1:29:33.120)

Then we had this tremendous challenge with the model itself.

Lex Fridman (1:29:36.160)

So now most people were pirating.

Lex Fridman (1:29:39.520)

But for the people who bought a download or a CD, the artists would get all the revenue

Lex Fridman (1:29:45.120)

for all the future plays then, right?

Lex Fridman (1:29:48.000)

So you got it all up front, whereas the streaming model was like almost nothing day one, almost

Lex Fridman (1:29:51.840)

nothing day two.

Lex Fridman (1:29:52.800)

And then at some point, this curve of incremental revenue would intersect with your day one

Lex Fridman (1:29:58.720)

payment.

Lex Fridman (1:29:59.840)

And that took a long time to play out before before the music labels, they understood

Lex Fridman (1:30:05.280)

that.

Lex Fridman (1:30:05.780)

But on the artist side, it took a lot of time to understand that actually, if I have a big

Gustav Soderstrom (1:30:09.600)

hit that is going to be played for many years, this is a much better model because I get

Gustav Soderstrom (1:30:14.000)

paid based on how much people use the product, not how much they thought they would use it

Lex Fridman (1:30:18.000)

day one or so forth.

Lex Fridman (1:30:20.080)

So it was a complicated model to get across.

Lex Fridman (1:30:22.880)

But time helped with that.

Lex Fridman (1:30:24.000)

And now the revenues to the music industry actually are bigger again than it's gone through

Lex Fridman (1:30:30.640)

this incredible dip and now they're back up.

Lex Fridman (1:30:32.000)

And so we're very proud of having been a part of that.

Lex Fridman (1:30:37.920)

So there have been distinct problems.

Gustav Soderstrom (1:30:39.520)

I think when it comes to the labels, we have taken the painful approach.

Gustav Soderstrom (1:30:46.720)

Some of our competition at the time, they kind of looked at other companies and said,

Gustav Soderstrom (1:30:52.400)

if we just ignore the rights, we get really big, really fast.

Lex Fridman (1:30:56.160)

We're going to be too big for the labels to kind of, too big to fail.

Gustav Soderstrom (1:31:00.480)

They're not going to kill us.

Lex Fridman (1:31:01.120)

We didn't take that approach.

Gustav Soderstrom (1:31:02.160)

We went legal from day one and we negotiated and negotiated and negotiated.

Lex Fridman (1:31:06.960)

It was very slow.

Gustav Soderstrom (1:31:07.600)

It was very frustrating.

Gustav Soderstrom (1:31:08.240)

We were angry at seeing other companies taking shortcuts and seeming to get away with it.

Gustav Soderstrom (1:31:12.800)

It was this game theory thing where over many rounds of playing the game, this would be

Lex Fridman (1:31:18.160)

the right strategy.

Lex Fridman (1:31:19.200)

And even though clearly there's a lot of frustrations at times during renegotiations, there is this

Lex Fridman (1:31:25.680)

there is this weird trust where we have been honest and fair.

Gustav Soderstrom (1:31:31.760)

We've never screwed them.

Lex Fridman (1:31:32.480)

They've never screwed us.

Gustav Soderstrom (1:31:33.680)

It's 10 years, but there's this trust and like they know that if music doesn't get

Gustav Soderstrom (1:31:39.280)

really big, if lots of people do not want to listen to music and want to pay for it,

Gustav Soderstrom (1:31:43.360)

Spotify has no business model.

Lex Fridman (1:31:44.960)

So we actually are incredibly aligned.

Gustav Soderstrom (1:31:48.240)

Other companies, not to be tense, but other companies have other business models where

Lex Fridman (1:31:51.840)

even if they made no money from music, they'd still be profitable companies.

Lex Fridman (1:31:56.400)

But Spotify won't.

Lex Fridman (1:31:57.200)

So I think the industry sees that we are actually aligned business wise.

Lex Fridman (1:32:03.120)

So there is this trust that allows us to do product development, even if it's scary,

Lex Fridman (1:32:11.040)

taking risks.

Gustav Soderstrom (1:32:12.560)

The free model itself was an incredible risk for the music industry to take that they should

Lex Fridman (1:32:17.200)

get credit for.

Gustav Soderstrom (1:32:17.920)

Now, some of it was that they had nothing to lose in the game.

Lex Fridman (1:32:20.400)

Some of it was that they had nothing to lose in Sweden.

Lex Fridman (1:32:22.240)

But frankly, a lot of the labels also took risk.

Lex Fridman (1:32:25.840)

And so I think we built up that trust with I think herding of cats sounds a bit.

Lex Fridman (1:32:32.320)

What's the word?

Lex Fridman (1:32:33.120)

It sounds like dismissive of the cats.

Gustav Soderstrom (1:32:35.280)

Dismissive.

Lex Fridman (1:32:35.920)

No, every cat matters.

Gustav Soderstrom (1:32:37.200)

They're all beautiful and very important.

Lex Fridman (1:32:39.360)

Exactly.

Gustav Soderstrom (1:32:39.920)

They've taken a lot of risks and certainly it's been frustrating.

Lex Fridman (1:32:44.960)

So it's really like playing it's game theory.

Gustav Soderstrom (1:32:47.600)

If you play the game many times, then you can have the statistical outcome that you

Lex Fridman (1:32:53.920)

bet on.

Lex Fridman (1:32:54.560)

And it feels very painful when you're in the middle of that thing.

Lex Fridman (1:32:57.520)

I mean, there's risk, there's trust, there's relationships.

Gustav Soderstrom (1:33:00.480)

From just having read the biography of Steve Jobs, similar kind of relationships were discussed

Lex Fridman (1:33:07.200)

in iTunes.

Gustav Soderstrom (1:33:08.400)

The idea of selling a song for a dollar was very uncomfortable for labels.

Lex Fridman (1:33:12.640)

Exactly.

Lex Fridman (1:33:13.760)

And there was no, it was the same kind of thing.

Gustav Soderstrom (1:33:16.400)

It was trust, it was game theory as a lot of relationships that had to be built.

Lex Fridman (1:33:21.840)

And it's really a terrifyingly difficult process that Apple could go through a little

Lex Fridman (1:33:28.880)

bit because they could afford for that process to fail.

Gustav Soderstrom (1:33:32.720)

For Spotify, it seems terrifying because you can't.

Gustav Soderstrom (1:33:37.600)

Initially, I think a lot of it comes down to honestly Daniel and his tenacity in negotiating,

Gustav Soderstrom (1:33:44.240)

which seems like an impossible task because he was completely unknown and so forth.

Lex Fridman (1:33:50.800)

But maybe that was also the reason that it worked.

Lex Fridman (1:33:56.480)

But I think game theory is probably the best way to think about it.

Gustav Soderstrom (1:34:03.120)

You could go straight for this Nash equilibrium that someone is going to defect or you play

Gustav Soderstrom (1:34:08.800)

it many times, you try to actually go for the top left, the corporations sell.

Lex Fridman (1:34:14.240)

Is there any magical reason why Spotify seems to have won this?

Lex Fridman (1:34:20.400)

So a lot of people have tried to do what Spotify tried to do and Spotify has come out.

Gustav Soderstrom (1:34:25.360)

Well, so the answer is that there's no magical reason because I don't believe in magic.

Lex Fridman (1:34:30.000)

But I think there are there are reasons.

Lex Fridman (1:34:32.240)

And I think some of them are that people have misunderstood a lot of what we actually do.

Gustav Soderstrom (1:34:40.400)

The actual Spotify model is very complicated.

Gustav Soderstrom (1:34:43.520)

They've looked at the premium model and said, it seems like you can charge $9.99 for music

Lex Fridman (1:34:49.200)

and people are going to pay, but that's not what happened.

Gustav Soderstrom (1:34:52.000)

Actually, when we launched the original mobile product, everyone said they would never pay.

Lex Fridman (1:34:56.640)

What happened was they started on the free product and then their engagement grew so

Lex Fridman (1:35:01.200)

much that eventually they said, maybe it is worth $9.99, right?

Gustav Soderstrom (1:35:05.680)

It's your propensity to pay gross with your engagement.

Lex Fridman (1:35:08.880)

So we have this super complicated business model.

Gustav Soderstrom (1:35:11.600)

We operate two different business models, advertising and premium at the same time.

Lex Fridman (1:35:15.760)

And I think that is hard to replicate.

Gustav Soderstrom (1:35:17.680)

I struggle to think of other companies that run large scale advertising and subscription

Lex Fridman (1:35:22.320)

products at the same time.

Lex Fridman (1:35:24.400)

So I think the business model is actually much more complicated than people think it is.

Lex Fridman (1:35:28.480)

And so some people went after just the premium part without the free part and ran into a

Gustav Soderstrom (1:35:32.800)

wall where no one wanted to pay.

Gustav Soderstrom (1:35:35.120)

Some people went after just music should be free, just ads, which doesn't give you enough

Gustav Soderstrom (1:35:40.400)

revenue and doesn't work for the music industry.

Lex Fridman (1:35:42.880)

So I think that combination is kind of opaque from the outside.

Lex Fridman (1:35:46.560)

So maybe I shouldn't say it here and reveal the secret, but that turns out to be hard

Lex Fridman (1:35:51.040)

to replicate than you would think.

Lex Fridman (1:35:54.400)

So there's a lot of brilliant business strategies out there.

Lex Fridman (1:35:57.040)

Brilliant business strategy here.

Lex Fridman (1:36:00.240)

Brilliance or luck?

Lex Fridman (1:36:01.280)

Probably more luck, but it doesn't really matter.

Gustav Soderstrom (1:36:03.520)

It looks brilliant in retrospect.

Lex Fridman (1:36:05.440)

Let's call it brilliant.

Gustav Soderstrom (1:36:07.840)

Yeah, when the books are written, they'll be brilliant.

Lex Fridman (1:36:10.480)

You've mentioned that your philosophy is to embrace change.

Lex Fridman (1:36:16.720)

So how will the music streaming and music listening world change over the next 10 years,

Lex Fridman (1:36:23.600)

20 years?

Gustav Soderstrom (1:36:24.640)

You look out into the far future.

Lex Fridman (1:36:26.960)

What do you think?

Gustav Soderstrom (1:36:28.960)

I think that music and for that matter, audio podcasts, audiobooks, I think it's one of

Lex Fridman (1:36:35.200)

the few core human needs.

Gustav Soderstrom (1:36:37.360)

I think it there is no good reason to me why it shouldn't be at the scale of something

Lex Fridman (1:36:41.680)

like messaging or social networking.

Gustav Soderstrom (1:36:44.160)

I don't think it's a niche thing to listen to music or news or something.

Lex Fridman (1:36:48.160)

So I think scale is obviously one of the things that I really hope for.

Gustav Soderstrom (1:36:50.880)

I think I hope that it's going to be billions of users.

Gustav Soderstrom (1:36:54.400)

I hope eventually everyone in the world gets access to all the world's music ever made.

Lex Fridman (1:36:58.720)

So obviously, I think it's going to be a much bigger business.

Lex Fridman (1:37:01.120)

Otherwise, we wouldn't be betting this big.

Gustav Soderstrom (1:37:05.040)

Now, if you look more at how it is consumed, what I'm hoping is back to this analogy of

Gustav Soderstrom (1:37:13.600)

the software tool chain, where I think I sometimes internally I make this analogy to text messaging.

Gustav Soderstrom (1:37:22.800)

Text messaging was also based on standards in the area of mobile carriers.

Lex Fridman (1:37:28.480)

You had the SMS, the 140 character, 120 character SMS.

Lex Fridman (1:37:33.600)

And it was great because everyone agreed on the standards.

Lex Fridman (1:37:36.080)

So as a consumer, you got a lot of distributions and interoperability, but it was a very constrained

Gustav Soderstrom (1:37:40.480)

format.

Lex Fridman (1:37:41.680)

And when the industry wanted to add pictures to that format to do the MMS, I looked it

Gustav Soderstrom (1:37:45.840)

up and I think it took from the late 80s to early 2000s.

Lex Fridman (1:37:48.720)

This is like a 15, 20 year product cycle to bring pictures into that.

Gustav Soderstrom (1:37:53.920)

Now, once that entire value chain of creation and consumption got wrapped in one software

Gustav Soderstrom (1:38:00.240)

stack within something like Snapchat or WhatsApp, the first week they added disappearing messages.

Gustav Soderstrom (1:38:07.280)

Then two weeks later, they added stories.

Gustav Soderstrom (1:38:09.600)

The pace of innovation when you're on one software stack and you can affect both creation

Lex Fridman (1:38:14.560)

and consumption, I think it's going to be rapid.

Lex Fridman (1:38:17.120)

So with these streaming services, we now, for the first time in history, have enough,

Gustav Soderstrom (1:38:22.320)

I hope, people on one of these services.

Gustav Soderstrom (1:38:25.040)

Actually, whether it's Spotify or Amazon or Apple or YouTube, and hopefully enough

Gustav Soderstrom (1:38:29.600)

creators that you can actually start working with the format again.

Lex Fridman (1:38:32.320)

And that excites me.

Gustav Soderstrom (1:38:33.760)

I think being able to change these constraints from 100 years, that could really do something

Lex Fridman (1:38:39.200)

interesting.

Gustav Soderstrom (1:38:40.160)

I really hope it's not just going to be the iteration on the same thing for the next 10

Lex Fridman (1:38:45.680)

to 20 years as well.

Gustav Soderstrom (1:38:47.360)

Yeah, changing the creation of music, the creation of audio, the creation of podcasts

Lex Fridman (1:38:52.000)

is a really fascinating possibility.

Gustav Soderstrom (1:38:54.400)

I myself don't understand what it is about podcasts that's so intimate.

Lex Fridman (1:38:59.680)

It just is.

Gustav Soderstrom (1:39:00.480)

I listen to a lot of podcasts.

Gustav Soderstrom (1:39:01.840)

I think it touches on a deep human need for connection that people do feel like they're

Gustav Soderstrom (1:39:09.680)

connected to when they listen.

Gustav Soderstrom (1:39:12.960)

I don't understand what the psychology of that is, but in this world that's becoming

Gustav Soderstrom (1:39:17.600)

more and more disconnected, it feels like this is fulfilling a certain kind of need.

Lex Fridman (1:39:24.800)

And empowering the creator as opposed to just the listener is really interesting.

Gustav Soderstrom (1:39:32.480)

I'm really excited that you're working on this.

Gustav Soderstrom (1:39:34.240)

Yeah, I think one of the things that is inspiring for our teams to work on podcasts is exactly

Gustav Soderstrom (1:39:38.800)

that, whether you think, like I probably do, that it's something biological about perceiving

Gustav Soderstrom (1:39:44.720)

to be in the middle of the conversation that makes you listen in a different way.

Gustav Soderstrom (1:39:47.840)

It doesn't really matter.

Lex Fridman (1:39:48.640)

People seem to perceive it differently.

Lex Fridman (1:39:50.240)

And there was this narrative for a long time that if you look at video, everything kind

Gustav Soderstrom (1:39:55.600)

of in the foreground, it got shorter and shorter and shorter because of financial pressures

Lex Fridman (1:39:59.840)

and monetization and so forth.

Lex Fridman (1:40:01.600)

And eventually, at the end, there's almost like 20 seconds clip, people just screaming

Gustav Soderstrom (1:40:06.240)

something and I feel really good about the fact that you could have interpreted that

Lex Fridman (1:40:14.640)

as people have no attention span anymore.

Gustav Soderstrom (1:40:16.880)

They don't want to listen to things.

Lex Fridman (1:40:18.400)

They're not interested in deeper stories.

Gustav Soderstrom (1:40:22.000)

People are getting dumber.

Lex Fridman (1:40:23.280)

But then podcasts came along and it's almost like, no, no, the need still existed.

Lex Fridman (1:40:28.000)

But maybe it was the fact that you're not prepared to look at your phone like this for

Lex Fridman (1:40:32.240)

two hours.

Lex Fridman (1:40:32.740)

But if you can drive at the same time, it seems like people really want to dig deeper

Lex Fridman (1:40:36.500)

and they want to hear like the more complicated version.

Lex Fridman (1:40:38.820)

So to me, that is very inspiring that that podcast is actually long form.

Gustav Soderstrom (1:40:42.980)

It gives me a lot of hope for humanity that people seem really interested in hearing deeper,

Gustav Soderstrom (1:40:48.340)

more complicated conversations.

Lex Fridman (1:40:49.940)

This is I don't understand it.

Gustav Soderstrom (1:40:52.100)

It's fascinating.

Lex Fridman (1:40:53.140)

So the majority for this podcast, listen to the whole thing.

Gustav Soderstrom (1:40:57.620)

This whole conversation we've been talking for an hour and 45 minutes.

Lex Fridman (1:41:02.500)

And somebody will I mean, most people will be listening to these words I'm speaking right

Gustav Soderstrom (1:41:06.580)

now.

Lex Fridman (1:41:06.580)

It's crazy.

Gustav Soderstrom (1:41:07.080)

You wouldn't have thought that 10 years ago with where the world seemed to go.

Lex Fridman (1:41:10.740)

That's very positive, I think.

Gustav Soderstrom (1:41:12.100)

That's really exciting.

Lex Fridman (1:41:13.300)

And empowering the creator there is really exciting.

Gustav Soderstrom (1:41:17.700)

Last question.

Lex Fridman (1:41:18.740)

You also have a passion for just mobile in general.

Lex Fridman (1:41:22.660)

How do you see the smartphone world, the digital space of smartphones and just everything that's

Gustav Soderstrom (1:41:32.660)

on the move, whether it's Internet of Things and so on, changing over the next 10 years

Lex Fridman (1:41:39.780)

and so on?

Gustav Soderstrom (1:41:41.460)

I think that one way to think about it is that computing might be moving out of these

Gustav Soderstrom (1:41:47.460)

multipurpose devices, the computer we had and the phone, into specific purpose devices.

Lex Fridman (1:41:55.140)

And it will be ambient that at least in my home, you just shout something at someone

Lex Fridman (1:42:01.060)

and there's always one of these speakers close enough.

Lex Fridman (1:42:03.380)

And so you start behaving differently.

Gustav Soderstrom (1:42:06.980)

It's as if you have the Internet ambient, ambiently around you and you can ask it things.

Lex Fridman (1:42:11.460)

So I think computing will kind of get more integrated and we won't necessarily think

Gustav Soderstrom (1:42:15.780)

of it as connected to a device in the same way that we do today.

Lex Fridman (1:42:21.700)

I don't know the path to that.

Gustav Soderstrom (1:42:22.900)

Maybe we used to have these desktop computers and then we partially replaced that with the

Lex Fridman (1:42:30.340)

laptops and left the desktop at home when I work.

Lex Fridman (1:42:32.740)

And then we got these phones and we started leaving the mobile phones.

Gustav Soderstrom (1:42:37.380)

We had the desktop at home when I work and then we got these phones and we started leaving

Gustav Soderstrom (1:42:41.540)

the laptop at home for a while.

Lex Fridman (1:42:42.820)

And maybe for stretches of time you're going to start using the watch and you can leave

Gustav Soderstrom (1:42:47.460)

your phone at home for a run or something.

Lex Fridman (1:42:50.580)

And we're on this progressive path where I think what is happening with voice is that

Gustav Soderstrom (1:43:00.740)

you have an interaction paradigm that doesn't require as large physical devices.

Lex Fridman (1:43:06.820)

So I definitely think there's a future where you can have your AirPods and your watch and

Gustav Soderstrom (1:43:12.820)

you can do a lot of computing.

Lex Fridman (1:43:15.860)

And I don't think it's going to be this binary thing.

Gustav Soderstrom (1:43:20.020)

I think it's going to be like many of us still have a laptop, we just use it less.

Lex Fridman (1:43:23.940)

And so you shift your consumption over.

Lex Fridman (1:43:26.820)

And I don't know about AR glasses and so forth.

Lex Fridman (1:43:31.940)

I'm excited about it.

Gustav Soderstrom (1:43:32.740)

I spent a lot of time in that area, but I still think it's quite far away.

Lex Fridman (1:43:35.700)

AR, VR, all of that.

Gustav Soderstrom (1:43:37.540)

Yeah, VR is happening and working.

Lex Fridman (1:43:39.780)

I think the recent Oculus Quest is quite impressive.

Gustav Soderstrom (1:43:43.940)

I think AR is further away.

Lex Fridman (1:43:45.300)

At least that type of AR.

Lex Fridman (1:43:48.100)

But I do think your phone or watch or glasses understanding where you are and maybe what

Lex Fridman (1:43:54.660)

you're looking at and being able to give you audio cues about that.

Lex Fridman (1:43:56.980)

Or you can say like, what is this?

Lex Fridman (1:43:58.580)

And it tells you what it is.

Gustav Soderstrom (1:44:00.980)

That I think might happen.

Lex Fridman (1:44:02.340)

You use your watch or your glasses as a mouse pointer on reality.

Gustav Soderstrom (1:44:08.020)

I think it might be a while before...

Lex Fridman (1:44:09.460)

I might be wrong.

Gustav Soderstrom (1:44:10.180)

I hope I'm wrong.

Gustav Soderstrom (1:44:10.820)

I think it might be a while before we walk around with these big lab glasses that project

Gustav Soderstrom (1:44:14.820)

things.

Lex Fridman (1:44:15.620)

I agree with you.

Gustav Soderstrom (1:44:16.820)

It's actually really difficult when you have to understand the physical world enough to

Lex Fridman (1:44:23.060)

project onto it.

Gustav Soderstrom (1:44:25.300)

I lied about the last question.

Gustav Soderstrom (1:44:26.740)

Go ahead, because I just thought of audio and my favorite topic, which is the movie

Gustav Soderstrom (1:44:32.660)

Her, do you think, whether it's part of Spotify or not, we'll have, I don't know if you've

Lex Fridman (1:44:41.140)

seen the movie Her.

Gustav Soderstrom (1:44:42.180)

Absolutely.

Lex Fridman (1:44:45.060)

And there, audio is the primary form of interaction and the connection with another entity that

Gustav Soderstrom (1:44:53.300)

you can actually have a relationship with, that you fall in love with based on voice

Lex Fridman (1:44:59.300)

alone, audio alone.

Lex Fridman (1:45:00.740)

Do you think that's possible, first of all, based on audio alone to fall in love with

Lex Fridman (1:45:04.820)

somebody?

Gustav Soderstrom (1:45:05.380)

Somebody or...

Lex Fridman (1:45:06.580)

Well, yeah, let's go with somebody.

Gustav Soderstrom (1:45:08.020)

Just have a relationship based on audio alone.

Lex Fridman (1:45:11.700)

And second question to that, can we create an artificial intelligence system that allows

Lex Fridman (1:45:18.500)

one to fall in love with it and her, him with you?

Lex Fridman (1:45:21.940)

So this is my personal answer, speaking for me as a person, the answer is quite unequivocally

Gustav Soderstrom (1:45:29.940)

yes on both.

Gustav Soderstrom (1:45:32.820)

I think what we just said about podcasts and the feeling of being in the middle of a

Gustav Soderstrom (1:45:36.580)

conversation, if you could have an assistant where, and we just said that feels like a

Lex Fridman (1:45:42.660)

very personal setting.

Lex Fridman (1:45:43.940)

So if you walk around with these headphones and this thing, you're speaking with this

Lex Fridman (1:45:47.380)

thing all of the time that feels like it's in your brain.

Gustav Soderstrom (1:45:49.940)

I think it's going to be much easier to fall in love with than something that would be

Lex Fridman (1:45:53.700)

on your screen.

Gustav Soderstrom (1:45:54.740)

I think that's entirely possible.

Lex Fridman (1:45:56.340)

And then from the, you can probably answer this better than me, but from the concept

Gustav Soderstrom (1:46:00.500)

of if it's going to be possible to build a machine that can achieve that, I think whether

Gustav Soderstrom (1:46:07.060)

you think of it as, if you can fake it, the philosophical zombie that assimilates it enough

Gustav Soderstrom (1:46:12.740)

or it somehow actually is, I think there's, it's only a question.

Lex Fridman (1:46:17.700)

It's only a question if you ask me about time, I'd have a different answer.

Lex Fridman (1:46:20.500)

But if you say I've given some half infinite time, absolutely.

Lex Fridman (1:46:24.580)

I think it's just atoms and arrangement of information.

Gustav Soderstrom (1:46:29.620)

Well, I personally think that love is a lot simpler than people think.

Lex Fridman (1:46:33.780)

So we started with true romance and ended in love.

Gustav Soderstrom (1:46:37.780)

I don't see a better place to end.

Lex Fridman (1:46:39.780)

Beautiful.

Gustav Soderstrom (1:46:40.340)

Gustav, thanks so much for talking today.

Lex Fridman (1:46:41.860)

Thank you so much.

Gustav Soderstrom (1:46:42.420)

It was a lot of fun.

Lex Fridman (1:46:43.140)

It was fun.

Lex Fridman (20:02.480)

And the first thing you need to do is obviously to lower the price to free and then you need

Lex Fridman (20:07.600)

to be better somehow.

Lex Fridman (20:09.440)

And the way that Spotify was better was on the user experience, on the actual performance,

Gustav Soderstrom (20:15.040)

the latency of, you know, even if you had high bandwidth broadband, it would still take

Gustav Soderstrom (20:24.640)

you 30 seconds to a minute to download one of these tracks.

Lex Fridman (20:30.800)

So the Spotify experience of starting within the perceptual limit of immediacy, about 250

Gustav Soderstrom (20:35.360)

milliseconds, meant that the whole trick was it felt as if you had downloaded all of Pirate

Lex Fridman (20:41.520)

Bay.

Gustav Soderstrom (20:41.680)

It was on your hard drive.

Lex Fridman (20:42.800)

It was that fast, even though it wasn't.

Lex Fridman (20:45.360)

And it was still free.

Lex Fridman (20:46.720)

But somehow you were actually still being a legal citizen.

Lex Fridman (20:50.400)

And that was the trick that Spotify managed to pull off.

Lex Fridman (20:54.880)

So I've actually heard you say this or write this.

Lex Fridman (20:58.240)

And I was surprised that I wasn't aware of it because I just took it for granted.

Gustav Soderstrom (21:02.400)

You know, whenever an awesome thing comes along, you're just like, of course, it has

Gustav Soderstrom (21:05.920)

to be this way.

Lex Fridman (21:07.360)

That's exactly right.

Gustav Soderstrom (21:08.560)

That it felt like the entire world's libraries at my fingertips because of that latency being

Lex Fridman (21:14.720)

reduced.

Lex Fridman (21:15.440)

What was the technical challenge in reducing the latency?

Lex Fridman (21:18.640)

So there was a group of really, really talented engineers, one of them called Ludwig Strigius.

Gustav Soderstrom (21:25.280)

He wrote the, actually from Gothenburg, he wrote the initial, the uTorrent client, which

Gustav Soderstrom (21:32.080)

is kind of an interesting backstory to Spotify, that we have one of the top developers from

Gustav Soderstrom (21:38.480)

uTorrent clients as well.

Lex Fridman (21:39.840)

So he wrote uTorrent, the world's smallest uTorrent client.

Lex Fridman (21:42.320)

And then he was acquired very early by Daniel and Martin, who founded Spotify, and they

Lex Fridman (21:49.440)

actually sold the uTorrent client to BitTorrent, but kept Ludwig.

Lex Fridman (21:53.040)

So Spotify had a lot of experience within peer to peer networking.

Lex Fridman (21:59.040)

So the original innovation was a distribution innovation, where Spotify built an end to

Gustav Soderstrom (22:04.560)

end media distribution system up until only a few years ago, we actually hosted all the

Lex Fridman (22:08.160)

music ourselves.

Lex Fridman (22:09.440)

So we had both the service side and the client, and that meant that we could do things such

Gustav Soderstrom (22:13.360)

as having a peer to peer solution to use local caching on the client side, because back then

Gustav Soderstrom (22:19.200)

the world was mostly desktop.

Lex Fridman (22:20.800)

But we could also do things like hack the TCP protocols, things like Nagel's algorithm

Gustav Soderstrom (22:26.240)

for kind of exponential back off, or ramp up and just go full throttle and optimize

Lex Fridman (22:31.200)

for latency at the cost of bandwidth.

Lex Fridman (22:33.760)

And all of this end to end control meant that we could do an experience that felt like a

Lex Fridman (22:39.200)

step change.

Gustav Soderstrom (22:40.480)

These days, we actually are on GCP, we don't host our own stuff, and everyone is really

Lex Fridman (22:46.720)

fast these days.

Lex Fridman (22:47.360)

So that was the initial competitive advantage.

Lex Fridman (22:49.440)

But then obviously, you have to move on over time.

Lex Fridman (22:51.440)

And that was over 10 years ago, right?

Lex Fridman (22:54.480)

That was in 2008.

Gustav Soderstrom (22:55.840)

The product was launched in Sweden.

Lex Fridman (22:57.520)

It was in a beta, I think, 2007.

Lex Fridman (22:59.440)

And it was on the desktop, right?

Lex Fridman (23:00.800)

It was desktop only.

Gustav Soderstrom (23:01.840)

There's no phone.

Lex Fridman (23:03.840)

There was no phone.

Gustav Soderstrom (23:04.480)

The iPhone came out in 2008.

Lex Fridman (23:07.920)

But the App Store came out one year later, I think.

Lex Fridman (23:10.480)

So the writing was on the wall, but there was no phone yet.

Gustav Soderstrom (23:14.160)

You've mentioned that people would use Spotify to discover the songs they like, and then

Gustav Soderstrom (23:19.680)

they would torrent those songs to so they can copy it to their phone.

Lex Fridman (23:24.880)

Just hilarious.

Gustav Soderstrom (23:25.840)

Exactly.

Lex Fridman (23:26.320)

Not torrent, pirate.

Gustav Soderstrom (23:27.440)

Seriously, piracy does seem to be like a good guide for business models.

Lex Fridman (23:33.520)

Video content.

Gustav Soderstrom (23:34.560)

As far as I know, Spotify doesn't have video content.

Lex Fridman (23:37.600)

Well, we do have music videos, and we do have videos on the service.

Lex Fridman (23:42.080)

But the way we think about ourselves is that we're an audio service, and we think that

Gustav Soderstrom (23:48.320)

if you look at the amount of time that people spend on audio, it's actually very similar

Gustav Soderstrom (23:52.800)

to the amount of time that people spend on music.

Lex Fridman (23:55.200)

It's very similar to the amount of time that people spend on video.

Lex Fridman (23:58.640)

So the opportunity should be equally big.

Lex Fridman (24:02.000)

But today, it's not at all valued.

Gustav Soderstrom (24:03.520)

Videos value much higher.

Lex Fridman (24:05.040)

So we think it's basically completely undervalued.

Lex Fridman (24:08.320)

So we think of ourselves as an audio service.

Lex Fridman (24:10.560)

But within that audio service, I think video can make a lot of sense.

Gustav Soderstrom (24:14.000)

I think when you're discovering an artist, you probably do want to see them and understand

Lex Fridman (24:19.040)

who they are, to understand their identity.

Gustav Soderstrom (24:21.200)

You won't see that video every time.

Lex Fridman (24:22.400)

90% of the time, the phone is going to be in your pocket.

Gustav Soderstrom (24:25.120)

For podcasters, you use video.

Lex Fridman (24:27.280)

I think that can make a ton of sense.

Lex Fridman (24:28.560)

So we do have video, but we're an audio service where, think of it as we call it internally,

Lex Fridman (24:33.600)

backgroundable video.

Gustav Soderstrom (24:35.120)

Video that is helpful, but isn't the driver of the narrative.

Gustav Soderstrom (24:39.440)

I think also, if we look at YouTube, there's quite a few folks who listen to music on YouTube.

Lex Fridman (24:48.560)

So in some sense, YouTube is a bit of a competitor to Spotify, which is very strange to me that

Lex Fridman (24:55.280)

people use YouTube to listen to music.

Lex Fridman (24:57.920)

They play essentially the music videos, right?

Lex Fridman (25:00.640)

But don't watch the videos and put it in their pocket.

Gustav Soderstrom (25:03.360)

Well, I think it's similar to what, strangely, maybe it's similar to what we were for the

Gustav Soderstrom (25:12.240)

piracy networks, where YouTube, for historical reasons, have a lot of music videos.

Lex Fridman (25:20.640)

So people use YouTube for a lot of the discovery part of the process, I think.

Lex Fridman (25:25.040)

But then it's not a really good sort of, quote unquote, MP3 player, because it doesn't even

Gustav Soderstrom (25:29.520)

background.

Lex Fridman (25:29.920)

Then you have to keep the app in the foreground.

Lex Fridman (25:31.600)

So it's not a good consumption tool, but it's a decently good discovery.

Lex Fridman (25:36.160)

I mean, I think YouTube is a fantastic product.

Lex Fridman (25:38.400)

And I use it for all kinds of purposes.

Lex Fridman (25:40.320)

That's true.

Gustav Soderstrom (25:41.040)

If I were to admit something, I do use YouTube a little bit to assist in the discovery process

Lex Fridman (25:46.560)

of songs.

Lex Fridman (25:47.280)

And then if I like it, I'll add it to Spotify.

Lex Fridman (25:50.320)

But that's OK.

Gustav Soderstrom (25:51.760)

That's OK with us.

Lex Fridman (25:53.600)

OK, so sorry, we're jumping around a little bit.

Lex Fridman (25:55.520)

So it's kind of incredible.

Lex Fridman (25:58.560)

You look at Napster, you look at the early days of Spotify.

Lex Fridman (26:03.440)

One fascinating point is how do you grow a user base?

Lex Fridman (26:06.640)

So you're there in Sweden.

Gustav Soderstrom (26:08.960)

You have an idea.

Lex Fridman (26:10.320)

I saw the initial sketches that look terrible.

Lex Fridman (26:14.160)

How do you grow a user base from a few folks to millions?

Lex Fridman (26:19.280)

I think there are a bunch of tactical answers.

Lex Fridman (26:22.240)

So first of all, I think you need a great product.

Lex Fridman (26:24.160)

I don't think you take a bad product and market it to be successful.

Lex Fridman (26:30.080)

So you need a great product.

Lex Fridman (26:31.120)

But sorry to interrupt, but it's a totally new way to listen to music, too.

Lex Fridman (26:34.720)

So it's not just did people realize immediately that Spotify is a great product?

Lex Fridman (26:38.560)

No, I think they did.

Lex Fridman (26:40.240)

So back to the point of piracy, it was a totally new way to listen to music legally.

Lex Fridman (26:45.840)

But people had been used to the access model in Sweden

Lex Fridman (26:48.960)

and the rest of the world for a long time through piracy.

Lex Fridman (26:50.880)

So one way to think about Spotify, it was just legal and fast piracy.

Lex Fridman (26:54.720)

And so people have been using it for a long time.

Lex Fridman (26:56.960)

So they weren't alien to it.

Gustav Soderstrom (26:59.040)

They didn't really understand how it could be illegal

Lex Fridman (27:01.360)

because it seemed too fast and too good to be true,

Gustav Soderstrom (27:03.920)

which I think is a great product proposition if you can be too good to be true.

Lex Fridman (27:06.960)

But what I saw again and again was people showing each other,

Lex Fridman (27:09.760)

clicking the song, showing how fast it started and say, can you believe this?

Lex Fridman (27:13.200)

So I really think it was about speed.

Gustav Soderstrom (27:16.320)

Then we also had an invite program that was really meant for scaling

Lex Fridman (27:22.000)

because we hosted our own service.

Gustav Soderstrom (27:23.280)

We needed to control scaling.

Lex Fridman (27:25.040)

But that built a lot of expectation.

Lex Fridman (27:27.600)

And I don't want to say hype because hype implies that it wasn't true.

Gustav Soderstrom (27:32.880)

Excitement around the product. And we've replicated that when we launched in the US.

Gustav Soderstrom (27:38.560)

We also built up an invite only program first.

Gustav Soderstrom (27:41.200)

There are lots of tactics, but I think you need a great product to solve some problem.

Lex Fridman (27:46.160)

And basically the key innovation, there was technology,

Lex Fridman (27:51.440)

but on a meta level, the innovation was really the access model versus the ownership model.

Lex Fridman (27:55.600)

And that was tricky.

Lex Fridman (27:56.880)

A lot of people said that they wanted to be able to do it.

Gustav Soderstrom (28:01.440)

I mean, they wanted to own their music.

Lex Fridman (28:04.480)

They would never kind of rent it or borrow it.

Lex Fridman (28:07.520)

But I think the fact that we had a free tier,

Gustav Soderstrom (28:09.120)

which meant that you get to keep this music for life as well, helped quite a lot.

Lex Fridman (28:14.560)

So this is an interesting psychological point that maybe you can speak to.

Lex Fridman (28:18.560)

It was a big shift for me.

Gustav Soderstrom (28:22.240)

It's almost like I had to go to therapy for this.

Lex Fridman (28:26.240)

I think I would describe my early listening experience,

Lex Fridman (28:29.360)

and I think a lot of my friends do, as basically hoarding music.

Lex Fridman (28:33.280)

As you're like slowly, one song by one song,

Gustav Soderstrom (28:35.920)

or maybe albums, gathering a collection of music that you love.

Lex Fridman (28:40.960)

And you own it.

Gustav Soderstrom (28:42.080)

It's like often, especially with CDs or tape, you like physically had it.

Lex Fridman (28:46.960)

And what Spotify, what I had to come to grips with,

Gustav Soderstrom (28:50.240)

it was kind of liberating actually, is to throw away all the music.

Lex Fridman (28:55.520)

I've had this therapy session with lots of people.

Lex Fridman (28:58.480)

And I think the mental trick is, so actually we've seen the user data.

Lex Fridman (29:02.560)

When Spotify started, a lot of people did the exact same thing.

Gustav Soderstrom (29:05.040)

They started hoarding as if the music would disappear.

Lex Fridman (29:09.280)

Almost the equivalent of downloading.

Lex Fridman (29:10.880)

And so we had these playlists that had limits of like a few hundred thousand tracks.

Lex Fridman (29:16.080)

We figured no one will ever.

Gustav Soderstrom (29:17.360)

Well, they do.

Lex Fridman (29:18.560)

Nuts and hundreds and hundreds of thousands of tracks.

Lex Fridman (29:20.960)

And to this day, some people want to actually save, quote unquote,

Lex Fridman (29:25.760)

and then play the entire catalog.

Lex Fridman (29:26.960)

But I think the therapy session goes something like instead of throwing away your music,

Lex Fridman (29:34.080)

if you took your files and you stored them in the locker at Google,

Gustav Soderstrom (29:38.720)

it'd be a streaming service.

Lex Fridman (29:39.680)

It's just that in that locker, you have all the world's music now for free.

Lex Fridman (29:42.720)

So instead of giving away your music, you got all the music.

Lex Fridman (29:45.520)

It's yours.

Gustav Soderstrom (29:46.720)

You could think of it as having a copy of the world's catalog there forever.

Lex Fridman (29:50.240)

So you actually got more music instead of less.

Gustav Soderstrom (29:52.720)

It's just that you just took that hard disk and you sent it to someone who stored it for you.

Lex Fridman (29:58.720)

And once you go through that mental journey, I'm like, it's still my files.

Gustav Soderstrom (30:01.440)

They're just over there.

Lex Fridman (30:02.560)

And I just have 40 million or 50 million or something now.

Gustav Soderstrom (30:05.520)

Then people are like, OK, that's good.

Lex Fridman (30:07.600)

The problem is, I think, because you paid us a subscription,

Gustav Soderstrom (30:11.840)

if we hadn't had the free tier where you would feel like,

Lex Fridman (30:14.000)

even if I don't want to pay anymore, I still get to keep them.

Gustav Soderstrom (30:17.120)

You keep your playlist forever.

Lex Fridman (30:18.480)

They don't disappear even though you stop paying.

Gustav Soderstrom (30:20.240)

I think that was really important.

Lex Fridman (30:21.760)

If we would have started as, you know, you can put in all this time,

Lex Fridman (30:25.440)

but if you stop paying, you lose all your work.

Gustav Soderstrom (30:27.280)

I think that would have been a big challenge and was the big challenge for a lot of our competitors.

Gustav Soderstrom (30:31.760)

That's another reason why I think the free tier is really important.

Lex Fridman (30:34.880)

That people need to feel the security, that the work they put in,

Gustav Soderstrom (30:37.600)

it will never disappear, even if they decide not to pay.

Lex Fridman (30:40.800)

I like how you put the work you put in.

Gustav Soderstrom (30:42.880)

I actually stopped even thinking of it that way.

Lex Fridman (30:44.480)

I just actually Spotify taught me to just enjoy music as opposed to.

Gustav Soderstrom (30:50.080)

As opposed to what I was doing before, which is like in an unhealthy way, hoarding music.

Lex Fridman (30:58.560)

Which I found that because I was doing that,

Gustav Soderstrom (31:01.280)

I was listening to a small selection of songs way too much to where I was getting sick of them.

Lex Fridman (31:07.520)

Whereas Spotify, the more liberating kind of approach is I was just enjoying.

Gustav Soderstrom (31:11.680)

Of course, I listened to Stairway to Heaven over and over,

Lex Fridman (31:13.920)

but because of the extra variety, I don't get as sick of them.

Gustav Soderstrom (31:18.240)

There's an interesting statistic I saw.

Lex Fridman (31:21.520)

So Spotify has, maybe you can correct me, but over 50 million songs, tracks,

Lex Fridman (31:27.600)

and over 3 billion playlists.

Lex Fridman (31:31.360)

So 50 million songs and 3 billion playlists.

Gustav Soderstrom (31:35.520)

60 times more playlist songs.

Lex Fridman (31:38.480)

What do you make of that?

Gustav Soderstrom (31:39.920)

Yeah.

Lex Fridman (31:40.160)

So the way I think about it is that from a statistician or machine learning point of view,

Gustav Soderstrom (31:48.320)

you have all these, if you want to think about reinforcement learning,

Lex Fridman (31:52.000)

you have this state space of all the tracks.

Gustav Soderstrom (31:54.320)

You can take different journeys through this world.

Lex Fridman (32:00.160)

I think of these as people helping themselves and each other,

Gustav Soderstrom (32:05.200)

creating interesting vectors through this space of tracks.

Lex Fridman (32:08.720)

And then it's not so surprising that across many tens of millions of atomic units,

Gustav Soderstrom (32:14.080)

there will be billions of paths that make sense.

Lex Fridman (32:17.280)

And we're probably pretty quite far away from having found all of them.

Lex Fridman (32:21.920)

So kind of our job now is users, when Spotify started,

Lex Fridman (32:26.640)

it was really a search box that was for the time pretty powerful.

Lex Fridman (32:30.000)

And then I'd like to refer to it as this programming language called playlisting,

Lex Fridman (32:34.400)

where if you, as you probably were pretty good at music,

Gustav Soderstrom (32:36.800)

you knew your new releases, you knew your back catalog,

Lex Fridman (32:39.120)

you knew your star with the heaven,

Gustav Soderstrom (32:40.480)

you could create a soundtrack for yourself using this playlisting tool,

Lex Fridman (32:43.200)

this like meta programming language for music to soundtrack your life.

Lex Fridman (32:47.360)

And people who were good at music, it's back to how do you scale the product.

Lex Fridman (32:50.960)

For people who are good at music, that wasn't actually enough.

Gustav Soderstrom (32:53.760)

If you had the catalog and a good search tool,

Lex Fridman (32:55.840)

and you can create your own sessions,

Gustav Soderstrom (32:57.120)

you could create really good a soundtrack for your entire life.

Lex Fridman (33:01.120)

Probably perfectly personalized because you did it yourself.

Lex Fridman (33:04.000)

But the problem was most people, many people aren't that good at music.

Lex Fridman (33:06.880)

They just can't spend the time.

Gustav Soderstrom (33:08.480)

Even if you're very good at music, it's going to be hard to keep up.

Lex Fridman (33:10.800)

So what we did to try to scale this was to essentially try to build,

Gustav Soderstrom (33:16.400)

you can think of them as agents that this friend that some people had

Lex Fridman (33:20.480)

that helped them navigate this music catalog.

Gustav Soderstrom (33:22.800)

That's what we're trying to do for you.

Lex Fridman (33:24.800)

But also there is something like 200 million active users.

Gustav Soderstrom (33:32.640)

1 million active users on Spotify.

Lex Fridman (33:35.040)

So there it's okay.

Lex Fridman (33:36.640)

So from the machine learning perspective,

Lex Fridman (33:39.760)

you have these 200 million people plus they're creating.

Gustav Soderstrom (33:45.760)

It's really interesting to think of a playlist as,

Lex Fridman (33:51.760)

I mean, I don't know if you meant it that way,

Lex Fridman (33:53.200)

but it's almost like a programming language.

Lex Fridman (33:54.880)

It's or at least a trace of exploration of those individual agents.

Gustav Soderstrom (34:01.120)

The listeners and you have all this new tracks coming in.

Lex Fridman (34:06.000)

So it's a fascinating space that is ripe for machine learning.

Lex Fridman (34:11.680)

So is there, is it possible, how can playlists be used as data

Lex Fridman (34:18.080)

in terms of machine learning and to help Spotify organize the music?

Lex Fridman (34:24.160)

So we found in our data, not surprising that people who play listed lots

Lex Fridman (34:29.680)

they retain much better.

Gustav Soderstrom (34:30.720)

They had a great experience.

Lex Fridman (34:32.240)

And so our first attempt was to playlist for users.

Lex Fridman (34:35.920)

And so we acquired this company called Tunigo of editors and professional playlisters

Lex Fridman (34:41.360)

and kind of leveraged the maximum of human intelligence

Gustav Soderstrom (34:45.600)

to help build kind of these vectors through the track space for people.

Lex Fridman (34:52.480)

And that broadened the product.

Lex Fridman (34:54.320)

But then the obvious next, and we use statistical means,

Lex Fridman (34:57.840)

where they could see when they created a playlist, how did that playlist perform?

Gustav Soderstrom (35:02.080)

They could see skips of the songs, they could see how the songs perform,

Lex Fridman (35:04.800)

and they manually iterated the playlist to maximize performance for a large group of people.

Lex Fridman (35:10.720)

But there were never enough editors to playlists for you personally.

Lex Fridman (35:14.480)

So the promise of machine learning was to go from kind of group personalization

Gustav Soderstrom (35:18.240)

using editors and tools and statistics to individualization.

Lex Fridman (35:22.640)

And then what's so interesting about the 3 billion playlists we have is we ended,

Gustav Soderstrom (35:28.160)

the truth is we lucked out.

Lex Fridman (35:29.360)

This was not a priority strategy, as is often the case.

Gustav Soderstrom (35:32.880)

It looks really smart in hindsight, but it was dumb luck.

Lex Fridman (35:37.440)

We looked at these playlists and we had some people in the company,

Gustav Soderstrom (35:42.160)

a person named Eric Beranodson.

Lex Fridman (35:43.840)

He was really good at machine learning already back then in like 2007, 2008.

Gustav Soderstrom (35:48.560)

Back then it was mostly collaborative filtering and so forth.

Lex Fridman (35:51.600)

But we realized that what this is, is people are grouping tracks for themselves

Gustav Soderstrom (35:57.920)

that have some semantic meaning to them.

Lex Fridman (36:00.640)

And then they actually label it with a playlist name as well.

Lex Fridman (36:04.160)

So in a sense, people were grouping tracks along semantic dimensions and labeling them.

Lex Fridman (36:09.840)

And so could you use that information to find that latent embedding?

Lex Fridman (36:15.840)

And so we started playing around with collaborative filtering

Lex Fridman (36:20.960)

and we saw tremendous success with it.

Gustav Soderstrom (36:24.160)

Basically trying to extract some of these dimensions.

Lex Fridman (36:28.320)

And if you think about it, it's not surprising at all.

Gustav Soderstrom (36:30.880)

It'd be quite surprising if playlists were actually random,

Lex Fridman (36:34.880)

if they had no semantic meaning.

Gustav Soderstrom (36:36.880)

For most people, they group these tracks for some reason.

Lex Fridman (36:39.840)

So we just happened across this incredible data set.

Gustav Soderstrom (36:43.120)

Where people are taking these tens of millions of tracks

Lex Fridman (36:46.800)

and group them along different semantic vectors.

Lex Fridman (36:49.280)

And the semantics being outside the individual users.

Lex Fridman (36:52.720)

So it's some kind of universal.

Gustav Soderstrom (36:54.400)

There's a universal embedding that holds across people on this earth.

Gustav Soderstrom (36:59.760)

Yes, I do think that the embeddings you find are going to be reflective of the people who play listed.

Lex Fridman (37:05.440)

So if you have a lot of indie lovers who play list,

Lex Fridman (37:09.040)

your embedding is going to perform better there.

Lex Fridman (37:14.800)

But what we found was that yes, there were these latent similarities.

Lex Fridman (37:20.560)

They were very powerful.

Lex Fridman (37:22.000)

And it was interesting because I think that the people who play listed the most initially

Lex Fridman (37:28.720)

were the so called music aficionados who were really into music.

Lex Fridman (37:32.640)

And they often had a certain...

Lex Fridman (37:34.240)

Their taste was often geared towards a certain type of music.

Lex Fridman (37:38.800)

And so what surprised us, if you look at the problem from the outside,

Gustav Soderstrom (37:42.160)

you might expect that the algorithms would start performing best with mainstreamers first.

Gustav Soderstrom (37:47.840)

Because it somehow feels like an easier problem to solve mainstream taste

Lex Fridman (37:51.360)

than really particular taste.

Gustav Soderstrom (37:53.360)

It was the complete opposite for us.

Lex Fridman (37:55.120)

The recommendations performed fantastically for people who saw themselves as

Gustav Soderstrom (37:59.280)

having very unique taste.

Lex Fridman (38:00.960)

That's probably because all of them play listed.

Lex Fridman (38:03.280)

And they didn't perform so well for mainstreamers.

Lex Fridman (38:05.120)

They actually thought they were a bit too particular and unorthodox.

Lex Fridman (38:09.440)

So we had the complete opposite of what we expected.

Lex Fridman (38:12.000)

Success within the hardest problem first,

Lex Fridman (38:13.920)

and then had to try to scale to more mainstream recommendations.

Lex Fridman (38:17.600)

So you've also acquired Echo Nest that analyzes song data.

Lex Fridman (38:24.160)

So in your view, maybe you can talk about,

Lex Fridman (38:28.400)

so what kind of data is there from a machine learning perspective?

Gustav Soderstrom (38:31.680)

From a machine learning perspective, there's a huge amount.

Gustav Soderstrom (38:35.680)

We're talking about playlisting and just user data of what people are listening to,

Gustav Soderstrom (38:40.640)

the playlist they're constructing, and so on.

Lex Fridman (38:44.640)

And then there's the actual data within a song.

Lex Fridman (38:48.080)

What makes a song, I don't know, the actual waveforms.

Lex Fridman (38:54.160)

How do you mix the two?

Lex Fridman (38:55.680)

How much value is there in each?

Lex Fridman (38:57.200)

To me, it seems like user data is a romantic notion

Gustav Soderstrom (39:03.120)

that the song itself would contain useful information.

Lex Fridman (39:05.840)

But if I were to guess, user data would be much more powerful,

Gustav Soderstrom (39:09.840)

like playlists would be much more powerful.

Lex Fridman (39:11.840)

Yeah, so we use both.

Gustav Soderstrom (39:14.800)

Our biggest success initially was with playlist data

Lex Fridman (39:18.800)

without understanding anything about the structure of the song.

Lex Fridman (39:22.480)

But when we acquired Echo Nest, they had the inverse problem.

Lex Fridman (39:25.520)

They actually didn't have any play data.

Gustav Soderstrom (39:27.440)

They were just, they were a provider of recommendations,

Lex Fridman (39:29.680)

but they didn't actually have any play data.

Lex Fridman (39:31.840)

So they looked at the structure of songs, sonically,

Lex Fridman (39:36.640)

and they looked at Wikipedia for cultural references and so forth, right?

Lex Fridman (39:40.400)

And did a lot of NLU and so forth.

Lex Fridman (39:41.920)

So we got that skill into the company and combined kind of our user data

Gustav Soderstrom (39:47.600)

with their kind of content based.

Lex Fridman (39:51.600)

So you can think of it as we were user based

Lex Fridman (39:53.200)

and they were content based in their recommendations.

Lex Fridman (39:54.880)

And we combined those two.

Lex Fridman (39:56.960)

And for some cases where you have a new song that has no play data,

Lex Fridman (40:00.240)

obviously you have to try to go by either who the artist is

Gustav Soderstrom (40:04.960)

or the sonic information in the song or what it's similar to.

Lex Fridman (40:09.760)

So there's definitely a value in both and we do a lot in both,

Lex Fridman (40:12.720)

but I would say, yes, the user data captures things

Lex Fridman (40:16.080)

that have to do with culture in the greater society

Gustav Soderstrom (40:19.680)

that you would never see in the content itself.

Lex Fridman (40:23.440)

But that said, we have seen, we have a research lab in Paris

Gustav Soderstrom (40:28.880)

when we can talk more about that on machine learning on the creator side,

Lex Fridman (40:32.960)

what it can do for creators, not just for the consumers,

Lex Fridman (40:35.520)

but where we looked at how does the structure of a song

Lex Fridman (40:38.640)

actually affect the listening behavior?

Lex Fridman (40:40.800)

And it turns out that there is a lot of,

Lex Fridman (40:43.120)

we can predict things like skips based on the song itself.

Gustav Soderstrom (40:48.480)

We could say that maybe you should move that chorus a bit

Lex Fridman (40:50.880)

because your skip is going to go up here.

Gustav Soderstrom (40:52.720)

There is a lot of latent structure in the music,

Lex Fridman (40:54.400)

which is not surprising because it is some sort of mind hack.

Lex Fridman (40:58.640)

So there should be structure. That's probably what we respond to.

Lex Fridman (41:00.960)

You just blew my mind actually from the creator perspective.

Lex Fridman (41:05.520)

So that's a really interesting topic

Lex Fridman (41:08.000)

that probably most creators aren't taking advantage of, right?

Lex Fridman (41:11.920)

So I've recently got to interact with a few folks,

Lex Fridman (41:15.920)

YouTubers who are like obsessed with this idea of what do I do

Lex Fridman (41:24.320)

to make sure people keep watching the video?

Lex Fridman (41:27.840)

And they like look at the analytics of which point do people turn it off and so on.

Gustav Soderstrom (41:32.720)

First of all, I don't think that's healthy,

Lex Fridman (41:35.040)

but it's because you can do it a little too much.

Lex Fridman (41:38.320)

But it is a really powerful tool for helping the creative process.

Lex Fridman (41:42.240)

You just made me realize you could do the same thing for creation of music.

Lex Fridman (41:47.280)

And so is that something you've looked into?

Lex Fridman (41:51.360)

And can you speak to how much opportunity there is for that kind of thing?

Gustav Soderstrom (41:54.800)

Yeah, so I listened to the podcast with Ziraj and I thought it was fantastic

Lex Fridman (41:59.200)

and I reacted to the same thing where he said he posted something in the morning,

Gustav Soderstrom (42:04.160)

immediately watched the feedback where the drop off was

Lex Fridman (42:06.560)

and then responded to that in the afternoon,

Gustav Soderstrom (42:08.400)

which is quite different from how people make podcasts, for example.

Lex Fridman (42:12.080)

Yes, exactly.

Gustav Soderstrom (42:12.880)

I mean, the feedback loop is almost non existent.

Lex Fridman (42:15.040)

So if we back out one level, I think actually both for music and podcasts,

Gustav Soderstrom (42:21.120)

which we also do at Spotify,

Lex Fridman (42:23.600)

I think there's a tremendous opportunity just for the creation workflow.

Lex Fridman (42:27.440)

And I think it's really interesting speaking to you who,

Lex Fridman (42:30.960)

because you're a musician, a developer, and a podcaster.

Gustav Soderstrom (42:34.720)

If you think about those three different roles,

Lex Fridman (42:36.560)

if you make the leap as a musician,

Gustav Soderstrom (42:38.880)

if you think about it as a software tool chain, really,

Lex Fridman (42:42.960)

your DAW with the stems, that's the IDE, right?

Gustav Soderstrom (42:46.320)

That's where you work in source code format with what you're creating.

Lex Fridman (42:51.120)

Then you sit around and you play with that.

Lex Fridman (42:52.320)

And when you're happy, you compile that thing into some sort of AAC or MP3 or something.

Lex Fridman (42:57.520)

You do that because you get distribution.

Gustav Soderstrom (42:59.040)

There are so many runtimes for that MP3 across the world in car stairs and stuff.

Lex Fridman (43:02.240)

So if you kind of compile this execution,

Gustav Soderstrom (43:03.920)

you ship it out in kind of an old fashioned boxed software analogy.

Lex Fridman (43:09.280)

And then you hope for the best, right?

Lex Fridman (43:11.760)

But as a software developer, you would never do that.

Lex Fridman (43:16.080)

First, you go on GitHub and you collaborate with other creators.

Lex Fridman (43:19.440)

And then you think it'd be crazy to just ship one version of your software

Lex Fridman (43:22.800)

without doing an A B test, without any feedback loop.

Gustav Soderstrom (43:26.800)

Issue tracking.

Lex Fridman (43:28.320)

Exactly.

Lex Fridman (43:28.880)

And then you would look at the feedback loop and say,

Lex Fridman (43:31.760)

try to optimize that thing, right?

Lex Fridman (43:34.160)

So I think if you think of it as a very specific software tool chain,

Lex Fridman (43:38.880)

it looks quite arcane, the tools that a music creator has

Gustav Soderstrom (43:42.880)

versus what a software developer has.

Lex Fridman (43:45.360)

So that's kind of how we think about it.

Lex Fridman (43:48.400)

Why wouldn't a music creator have something like GitHub

Lex Fridman (43:52.640)

where you could collaborate much more easily?

Lex Fridman (43:54.000)

So we bought this company called Soundtrap,

Lex Fridman (43:56.560)

which has a kind of Google Docs for music approach, where you can collaborate

Gustav Soderstrom (44:01.680)

with other people on the kind of source code format with Stems.

Lex Fridman (44:05.600)

And I think introducing things like AI tools there to help you

Gustav Soderstrom (44:09.600)

as you're creating music, both in helping you put accompaniment to your music,

Lex Fridman (44:19.280)

like drums or something, help you master and mix automatically,

Gustav Soderstrom (44:24.400)

help you understand how this track will perform.

Lex Fridman (44:26.720)

Exactly what you would expect as a software developer.

Gustav Soderstrom (44:29.600)

I think it makes a lot of sense.

Lex Fridman (44:30.880)

And I think the same goes for a podcaster.

Gustav Soderstrom (44:33.520)

I think podcasters will expect to have the same kind of feedback loop

Lex Fridman (44:36.320)

that Siraj has, like, why wouldn't you?

Gustav Soderstrom (44:39.520)

Maybe it's not healthy, but...

Lex Fridman (44:41.520)

Sorry, I wanted to criticize the fact because you can overdo it

Gustav Soderstrom (44:45.120)

because a lot of the, and we're in a new era of that.

Lex Fridman (44:49.760)

So you can become addicted to it and therefore, what people say,

Gustav Soderstrom (44:56.400)

you become a slave to the YouTube algorithm or sort of,

Lex Fridman (45:00.640)

it's always a danger of a new technology as opposed to say,

Gustav Soderstrom (45:04.400)

if you're creating a song, becoming too obsessed about the intro riff to the song

Gustav Soderstrom (45:11.600)

that keeps people listening versus actually the entirety of the creation process.

Gustav Soderstrom (45:15.440)

It's a balance.

Lex Fridman (45:16.160)

But the fact that there's zero, I mean, you're blowing my mind right now,

Gustav Soderstrom (45:19.680)

because you're completely right that there is no signal whatsoever.

Lex Fridman (45:24.960)

There's no feedback whatsoever on the creation process and music or podcasting,

Gustav Soderstrom (45:30.000)

almost at all.

Lex Fridman (45:31.680)

And are you saying that Spotify is hoping to help create tools to, not tools, but...

Gustav Soderstrom (45:39.360)

No, tools actually.

Lex Fridman (45:41.680)

Actually, tools.

Gustav Soderstrom (45:42.640)

Tools for creators.

Lex Fridman (45:47.200)

Absolutely.

Lex Fridman (45:48.320)

So we've made some acquisitions the last few years around music creation,

Lex Fridman (45:53.520)

this company called Soundtrap, which is a digital audio workstation,

Lex Fridman (45:57.280)

but that is browser based.

Lex Fridman (45:59.040)

And their focus was really the Google Docs approach.

Gustav Soderstrom (46:01.200)

We can collaborate with people much more easily than you could in previous tools.

Lex Fridman (46:06.080)

So we have some of these tools that we're working with that we want to make accessible

Lex Fridman (46:09.280)

and then we can connect it with our consumption data.

Lex Fridman (46:12.960)

We can create this feedback loop where we could help you understand,

Gustav Soderstrom (46:16.800)

we could help you create and help you understand how you will perform.

Lex Fridman (46:20.960)

We also acquired this other company within podcasting called Anchor,

Gustav Soderstrom (46:24.560)

which is one of the biggest podcasting tools, mobile focused.

Lex Fridman (46:28.400)

So really focused on simple creation or easy access to creation.

Lex Fridman (46:32.800)

But that also gives us this feedback loop.

Lex Fridman (46:34.960)

And even before that, we invested in something called Spotify for Artists

Lex Fridman (46:40.640)

and Spotify for Podcasters, which is an app that you can download,

Lex Fridman (46:43.600)

you can verify that you are that creator.

Lex Fridman (46:46.000)

And then you get things that software developers have had for years.

Lex Fridman (46:51.680)

You can see where, if you look at your podcast, for example, on Spotify

Gustav Soderstrom (46:55.520)

or a song that you released, you can see how it's performing,

Lex Fridman (46:58.720)

which cities it's performing in, who's listening to it,

Gustav Soderstrom (47:01.280)

what's the demographic breakup.

Lex Fridman (47:02.800)

So similar in the sense that you can understand

Lex Fridman (47:05.840)

how you're actually doing on the platform.

Lex Fridman (47:08.880)

So we definitely want to build tools.

Gustav Soderstrom (47:10.480)

I think you also interviewed the head of research for Adobe.

Lex Fridman (47:15.920)

And I think that's an, back to Photoshop that you like,

Gustav Soderstrom (47:19.680)

I think that's an interesting analogy as well.

Gustav Soderstrom (47:22.800)

Photoshop, I think, has been very innovative in helping photographers and artists.

Lex Fridman (47:28.000)

And I think there should be the same kind of tools for music creators,

Lex Fridman (47:32.320)

where you could get AI assistance, for example, as you're creating music,

Gustav Soderstrom (47:36.640)

as you can do with Adobe, where you can,

Lex Fridman (47:38.880)

I want a sky over here and you can get help creating that sky.

Gustav Soderstrom (47:42.000)

The really fascinating thing is what Adobe doesn't have

Lex Fridman (47:47.520)

is a distribution for the content you create.

Lex Fridman (47:50.400)

So you don't have the data of if I create, if I, you know,

Lex Fridman (47:55.840)

whatever creation I make in Photoshop or Premiere,

Gustav Soderstrom (47:59.360)

I can't get like immediate feedback like I can on YouTube,

Lex Fridman (48:02.480)

for example, about the way people are responding.

Lex Fridman (48:05.360)

And if Spotify is creating those tools, that's a really exciting actually world.

Lex Fridman (48:11.680)

But let's talk a little about podcasts.

Lex Fridman (48:16.720)

So I have trouble talking to one person.

Lex Fridman (48:20.000)

So it's a bit terrifying and kind of hard to fathom,

Lex Fridman (48:23.120)

but on average, 60 to 100,000 people will listen to this episode.

Lex Fridman (48:30.320)

Okay, so it's intimidating.

Gustav Soderstrom (48:32.240)

Yeah, it's intimidating.

Lex Fridman (48:34.320)

So I hosted on Blueberry.

Gustav Soderstrom (48:36.720)

I don't know if I'm pronouncing that correctly, actually.

Lex Fridman (48:39.520)

It looks like most people listen to it on Apple Podcasts,

Gustav Soderstrom (48:42.400)

Cast Box and Pocket Casts, and only about a thousand listen on Spotify.

Lex Fridman (48:48.480)

It's just my podcast, right?

Lex Fridman (48:53.840)

So where do you see a time when Spotify will dominate this?

Lex Fridman (49:00.960)

So Spotify is relatively new into this podcasting site.

Gustav Soderstrom (49:06.000)

Yeah, in podcasting.

Lex Fridman (49:07.520)

What's the deal with podcasting and Spotify?

Lex Fridman (49:10.800)

How serious is Spotify about podcasting?

Lex Fridman (49:13.440)

Do you see a time where everybody would listen to, you know,

Lex Fridman (49:16.800)

probably a huge amount of people, majority perhaps listen to music on Spotify?

Lex Fridman (49:22.400)

Do you see a time when the same is true for podcasting?

Gustav Soderstrom (49:26.880)

Well, I certainly hope so.

Lex Fridman (49:28.560)

That is our mission.

Gustav Soderstrom (49:29.360)

Our mission as a company is actually to enable a million creators to live off of their art,

Lex Fridman (49:34.160)

and a billion people be inspired by it.

Lex Fridman (49:35.840)

And what I think is interesting about that mission is it actually puts the creators first,

Lex Fridman (49:40.640)

even though it started as a consumer focused company,

Lex Fridman (49:43.040)

and it's just to be able to live off of their art,

Lex Fridman (49:44.800)

not just make some money off of their art as well.

Lex Fridman (49:47.840)

So it's quite an ambitious project.

Lex Fridman (49:51.920)

So we think about creators of all kinds,

Lex Fridman (49:53.920)

and we kind of expanded our mission from being music to being audio a while back.

Lex Fridman (50:01.120)

And that's not so much because we think we made that decision.

Gustav Soderstrom (50:08.400)

We think that decision was made for us.

Lex Fridman (50:10.800)

We think the world made that decision.

Gustav Soderstrom (50:12.960)

Whether we like it or not, when you put in your headphones,

Gustav Soderstrom (50:16.560)

you're going to make a choice between music and a new episode of your podcast or something else.

Gustav Soderstrom (50:25.440)

We're in that world whether we like it or not.

Lex Fridman (50:26.960)

And that's how radio works.

Lex Fridman (50:28.960)

So we decided that we think it's about audio.

Lex Fridman (50:32.320)

You can see the rise of audiobooks and so forth.

Gustav Soderstrom (50:34.480)

We think audio is a great opportunity.

Lex Fridman (50:36.480)

So we decided to enter it.

Lex Fridman (50:37.600)

And obviously, Apple and Apple Podcasts is absolutely dominating in podcasting,

Lex Fridman (50:45.280)

and we didn't have a single podcast only like two years ago.

Lex Fridman (50:49.440)

What we did though was we looked at this and said,

Lex Fridman (50:54.560)

can we bring something to this?

Gustav Soderstrom (50:56.480)

We want to do this, but back to the original Spotify,

Lex Fridman (50:59.200)

we have to do something that consumers actually value to be able to do this.

Lex Fridman (51:03.840)

And the reason we've gone from not existing at all to being quite a wide margin,

Gustav Soderstrom (51:09.840)

the second largest podcast consumption, still wide gap to iTunes, but we're growing quite fast.

Gustav Soderstrom (51:16.480)

I think it's because when we looked at the consumer problem,

Gustav Soderstrom (51:20.320)

people said surprisingly that they wanted their podcasts and music in the same application.

Lex Fridman (51:26.960)

So what we did was we took a little bit of a different approach where we said,

Lex Fridman (51:29.760)

instead of building a separate podcast app,

Lex Fridman (51:31.440)

we thought, is there a consumer problem to solve here?

Lex Fridman (51:33.680)

Because the others are very successful already.

Lex Fridman (51:35.680)

And we thought there was in making a more seamless experience

Lex Fridman (51:38.960)

where you can have your podcast and your music in the same application,

Gustav Soderstrom (51:43.680)

because we think it's audio to you.

Lex Fridman (51:45.440)

And that has been successful.

Lex Fridman (51:46.800)

And that meant that we actually had 200 million people to offer this to instead of starting from zero.

Lex Fridman (51:52.400)

So I think we have a good chance because we're taking a different approach than the competition.

Lex Fridman (51:56.880)

And back to the other thing I mentioned about

Lex Fridman (51:59.120)

creators, because we're looking at the end to end flow.

Gustav Soderstrom (52:02.800)

I think there's a tremendous amount of innovation to do around podcast as a format.

Gustav Soderstrom (52:07.040)

When we have creation tools and consumption, I think we could start improving what podcasting is.

Gustav Soderstrom (52:12.640)

I mean, podcast is this opaque, big, like one, two hour file that you're streaming,

Lex Fridman (52:19.520)

which it really doesn't make that much sense in 2019 that it's not interactive.

Gustav Soderstrom (52:24.240)

There's no feedback loops, nothing like that.

Lex Fridman (52:26.000)

So I think if we're going to win, it's going to have to be because we build a better product

Gustav Soderstrom (52:29.760)

for creators and for consumers.

Lex Fridman (52:32.480)

So we'll see, but it's certainly our goal.

Gustav Soderstrom (52:34.640)

We have a long way to go.

Lex Fridman (52:36.240)

Well, the creators part is really exciting.

Gustav Soderstrom (52:38.160)

You already, you got me hooked there.

Lex Fridman (52:40.160)

Cause the only stats I have,

Gustav Soderstrom (52:42.320)

Blueberry just recently added the stats of whether it's listened to the end or not.

Lex Fridman (52:48.560)

And that's like a huge improvement, but that's still

Gustav Soderstrom (52:52.320)

nowhere to where you could possibly go in terms of statistics.

Lex Fridman (52:54.960)

You just download the Spotify podcasters up and verify.

Lex Fridman (52:57.200)

And then, then you'll know where people dropped out in this episode.

Lex Fridman (52:59.920)

Oh, wow.

Gustav Soderstrom (53:00.400)

Okay.

Lex Fridman (53:01.600)

The moment I started talking.

Gustav Soderstrom (53:02.800)

Okay.

Lex Fridman (53:03.360)

I might be depressed by this, but okay.

Lex Fridman (53:06.800)

So one, um, one other question is the original Spotify for music.

Lex Fridman (53:14.400)

And I have a question about podcasting in this line is the idea of podcasting

Gustav Soderstrom (53:19.120)

about podcasting in this line is the idea of albums.

Lex Fridman (53:23.440)

I have, uh, what did you, uh, music aficionados, uh, friends who are really,

Gustav Soderstrom (53:29.440)

uh, big fans of music often, uh, really enjoy albums,

Lex Fridman (53:33.280)

listening to entire albums of, of an artist.

Gustav Soderstrom (53:36.400)

Correct me if I'm wrong, but I feel like Spotify has helped

Lex Fridman (53:40.960)

replace the idea of an album with playlists.

Lex Fridman (53:44.240)

So you create your own albums.

Lex Fridman (53:46.000)

It's, it's kind of the way, at least I've experienced music

Lex Fridman (53:48.880)

and I've really enjoyed it that way.

Lex Fridman (53:51.040)

One of the things that was missing in podcasting for me,

Gustav Soderstrom (53:54.880)

I don't know if it's missing.

Lex Fridman (53:56.320)

I don't know.

Gustav Soderstrom (53:56.880)

It's an open question for me, but the way I listened to podcasts is

Lex Fridman (53:59.920)

the way I would listen to albums.

Lex Fridman (54:02.080)

So I take a Joe Rogan experience and that's an album.

Lex Fridman (54:05.600)

And I listened, you know, I like, I, I put that on and I listened one

Gustav Soderstrom (54:09.680)

episode after the next, then there's a sequence and so on.

Lex Fridman (54:12.640)

Is there a room for doing what you did for music or doing what

Gustav Soderstrom (54:17.520)

Spotify did for music, but, uh, creating playlists, sort of, uh,

Lex Fridman (54:22.880)

this kind of playlisting idea of breaking apart from podcasting,

Gustav Soderstrom (54:27.120)

uh, from individual podcasts and creating kind of, uh, this interplay

Lex Fridman (54:31.680)

or, or have you thought about that space?

Gustav Soderstrom (54:33.760)

Uh, it's a great question.

Lex Fridman (54:34.800)

So I think in, um, in music, you're right.

Gustav Soderstrom (54:38.720)

Basically you bought an album.

Lex Fridman (54:39.920)

So it was like, you bought a small catalog of like 10 tracks, right?

Gustav Soderstrom (54:42.800)

It was, it was, again, it was actually a lot of, a lot of consumption.

Lex Fridman (54:46.720)

You think it's about what you like, but it's based on the business model.

Lex Fridman (54:49.680)

So you paid for this 10 track service and then you listened to that for a while.

Lex Fridman (54:54.240)

And then when, when everything was flat priced, you tended to listen differently.

Gustav Soderstrom (54:58.480)

Now, so, so I think the, I think the album is still tremendously important.

Lex Fridman (55:01.360)

That's why we have it and you can save albums and so forth.

Lex Fridman (55:03.360)

And you have a huge amount of people who really listen according to albums.

Lex Fridman (55:06.480)

And I like that because it is a creator format, you can tell a longer story

Gustav Soderstrom (55:10.240)

over several tracks.

Lex Fridman (55:12.000)

And so some people listen to just one track.

Gustav Soderstrom (55:13.840)

Some people actually want to hear that whole story.

Lex Fridman (55:17.520)

Now in podcast, I think, I think it's different.

Gustav Soderstrom (55:21.600)

You can argue that podcasts might be more like shows on Netflix.

Lex Fridman (55:25.600)

Have like a full season of Narcos and you're probably not going to do like

Gustav Soderstrom (55:29.200)

one episode of Narcos and then one of House of Cards, like, like, you know,

Lex Fridman (55:33.440)

there's a narrative there.

Lex Fridman (55:34.480)

And you, you, you love the cast and you love these characters.

Lex Fridman (55:37.440)

So I think people will, people love shows.

Lex Fridman (55:42.000)

And I think they will, they will listen to those shows.

Lex Fridman (55:44.880)

I do think you follow a bunch of shows at the same time.

Lex Fridman (55:46.880)

So there's certainly an opportunity to bring you the latest episode of, you

Lex Fridman (55:50.480)

know, whatever the five, six, 10 things that, that you're into.

Gustav Soderstrom (55:54.560)

But, but I think, I think people are going to listen to specific hosts and love

Lex Fridman (56:00.000)

those hosts for a long time.

Gustav Soderstrom (56:01.600)

Because I think there's something different with podcasts where, um, this

Lex Fridman (56:06.880)

format of the, the, the, the, the, the experience of the, of the audience is

Gustav Soderstrom (56:11.280)

actually sitting here right between us.

Lex Fridman (56:13.360)

Whereas if you look at something on TV, the audio actually would come from, you

Gustav Soderstrom (56:16.960)

would sit over there and the audio would come to you from both of us as if you

Lex Fridman (56:20.080)

were watching, not as you were part of the conversation.

Lex Fridman (56:22.560)

So my experience is having listened to podcasts like yours and Joe Rogan is, I

Lex Fridman (56:27.280)

feel like I know all of these people.

Gustav Soderstrom (56:28.720)

They, they have a lot of experience.

Lex Fridman (56:30.240)

I know all of these people, they have no idea who I am, but I feel like I've

Gustav Soderstrom (56:33.600)

listened to so many hours of that.

Gustav Soderstrom (56:35.040)

It's very different from me watching a, watching like a TV show or an interview.

Lex Fridman (56:39.440)

So I think you, you kind of, um, fall in love with people and, um, experience

Lex Fridman (56:44.560)

in a, in a different way.

Lex Fridman (56:45.760)

So I think, I think shows and hosts are going to be very, uh, very important.

Lex Fridman (56:49.280)

I don't think that's going to go away into some sort of thing where, where you

Gustav Soderstrom (56:52.160)

don't even know who you're listening to.

Lex Fridman (56:53.360)

I don't think that's going to happen.

Lex Fridman (56:55.040)

What I do think is I think there's a tremendous discovery opportunity in

Lex Fridman (56:59.760)

podcast because the catalog is growing quite quickly.

Lex Fridman (57:03.920)

And I think podcast is only a few, like five, 600,000 shows right now.

Lex Fridman (57:11.360)

If you look back to YouTube as another analogy of creators, no one really knows

Gustav Soderstrom (57:16.080)

if you would lift the lid on YouTube, but it's probably billions of episodes.

Lex Fridman (57:21.120)

And so I think the podcast catalog would probably grow tremendously because the

Gustav Soderstrom (57:24.960)

creation tools are getting easier.

Lex Fridman (57:27.040)

And then you're going to have this discovery opportunity that I think is

Gustav Soderstrom (57:30.800)

really big.

Lex Fridman (57:31.280)

So, so a lot of people tell me that they love their shows, but discovering

Gustav Soderstrom (57:35.600)

podcasts kind of suck.

Lex Fridman (57:36.880)

It's really hard to get into new show.

Gustav Soderstrom (57:38.720)

They're usually quite long.

Lex Fridman (57:39.840)

It's a big time investment.

Lex Fridman (57:40.960)

So I think there's plenty of opportunity in the discovery part.

Lex Fridman (57:45.600)

Yeah, for sure.

Gustav Soderstrom (57:46.560)

A hundred percent in, in even the dumbest, there's so many low hanging fruit too.

Gustav Soderstrom (57:51.200)

Uh, for example, just knowing what episode to listen to first to try out a podcast.

Gustav Soderstrom (57:59.680)

Exactly.

Lex Fridman (58:00.400)

Uh, because most podcasts don't have an order to them.

Gustav Soderstrom (58:03.920)

Uh, they, they can be listened to out of order and sorry to say some are better

Lex Fridman (58:10.880)

than others episodes.

Lex Fridman (58:12.560)

So some episodes of Joe Rogan are better than others.

Lex Fridman (58:15.520)

And it's nice to know, uh, which you should listen to, to try it out.

Lex Fridman (58:20.400)

And there's, uh, as far as I know, almost no information, uh, in terms of like, uh,

Lex Fridman (58:26.320)

upvotes on how good an episode is.

Gustav Soderstrom (58:28.640)

Exactly.

Lex Fridman (58:29.280)

So I think part of the problem is, uh, you, it's kind of like music.

Gustav Soderstrom (58:33.520)

There isn't one answer.

Gustav Soderstrom (58:34.480)

People use music for different things and there's actually many different types of music.

Gustav Soderstrom (58:37.440)

There's workout music and there's classical piano music and focus music and,

Lex Fridman (58:41.200)

and, and, uh, so forth.

Gustav Soderstrom (58:42.640)

I think the same with podcasts.

Lex Fridman (58:44.080)

Some podcasts are sequential.

Gustav Soderstrom (58:45.360)

They're supposed to be listened to in, in order.

Lex Fridman (58:48.400)

It's actually, it's actually telling a narrative.

Gustav Soderstrom (58:51.040)

Some podcasts are one topic, uh, kind of like yours, but different guests.

Lex Fridman (58:55.840)

So you could jump in anywhere.

Gustav Soderstrom (58:57.280)

Some podcasts actually have completely different topics.

Lex Fridman (58:59.440)

And for those podcasts, it might be that I want, you know, we should recommend one episode

Gustav Soderstrom (59:04.560)

because it's about AI from someone, but then they talk about something that you're not

Lex Fridman (59:09.280)

interested in the rest of the episodes.

Lex Fridman (59:10.880)

So I think our, what we're spending a lot of time on now is just first understanding

Gustav Soderstrom (59:15.040)

the domain and creating kind of the knowledge graph of how do these objects relate and how

Gustav Soderstrom (59:21.520)

do people consume.

Lex Fridman (59:22.240)

And I think we'll find that it's going to be, it's going to be different.

Gustav Soderstrom (59:26.000)

I'm excited because you're the, uh, Spotify is the first people I'm aware of that are

Lex Fridman (59:32.240)

trying to do this for podcasting.

Gustav Soderstrom (59:34.800)

Podcasting has been like a wild west up until now.

Gustav Soderstrom (59:38.240)

It's been a very, we want to be very careful though, because it's been a very good wild

Gustav Soderstrom (59:43.120)

west, I think it's this fragile ecosystem.

Lex Fridman (59:46.320)

And I, we want to make sure that you don't barge in and say like, Oh, we're going to

Gustav Soderstrom (59:52.080)

internetize this thing.

Lex Fridman (59:53.440)

And you have to think about the creators.

Gustav Soderstrom (59:56.640)

You have to understand how they get distribution today, who listens to how they make money