Speaker unverified: prediction

Transcript context

…I love that. So I was listening to it, as I said, and what was really interesting is I think episode four was actually all about AI, and I think your first attempts at leveraging machine learning in AI within Spotify. And I think that's what led to Discover Weekly and a few other tools. And that was years ago. And it's interesting listening to it now where AI is, again, a huge deal. And so I'm curious very tactically on the product team what you advise product managers and product teams on how to think about AI in their product thinking and also just in their day-to-day work. Gustav Söderström I can give a few examples there and I don't know that we're more sophisticated than anyone else, but we'll be doing at least the traditional machine learning for quite a long time. And I think in the podcast, I think I talked about the journey of the internet in stages. And one way to think about it is that the internet started with curation of the user curation. So you took something, some good, like people or books or music and you digitize it and you put it online, and then you ask users to curate it. And that was your Facebook, Spotify and so forth. And then after a while, the world switched from curation to recommendation, where instead of people doing that work, you had algorithms. And that was a big change that required us and others to actually rethink the entire user experience and sometimes the business model as well. Gustav Söderström And I think what we're entering now is we're going from your curation to recommendation to generation. And I suspect it will be as big of a shift that you will eventually have to rethink your products. So that's one lens. So I tend to talk to my teams about, even though it's all machine learning, I ask them to think of this as something completely different. The recommendation error was one type of machine learning. The generation error is a different type, so don't think of it as just more of the same, think of it as something actually completely new instead. And what we learned in ... Well, a few things. So if you look at this new era of large language models and the fusion models and so forth, there are two types of applications. As I said for the recommendation error, we had to rethink the user interface and the experience for recommendation first error. Gustav Söderström And so what does that mean in the generative area? No one really knows yet. As usual, there are a bunch of iterative improvements. So we use these large language models to improve our recommendations. You can have bigger vectors that can have more cultural knowledge. You can use it for safety classification on podcasts that no one has listened to yet and so forth. So there's lots of obvious improvements and we're doing those. But so far, we've only really done one real generative product in the hard definition, which is a product that couldn't have existed without generative AI, and that is the AI DJ. So that's a concept that we've been thinking about for a very long time. And the AI DJ is you press a button, a digitized person, there's a real person named X, digitized X. So he's now an AI, comes on and talks to you about music that you like and suggests music, and you can listen to it. And if you don't like it, you can just call him back and he says, "Okay, now, let's listen to something maybe from a few summers ago," or "Here's some new stuff that were trending yesterday in The Last of Us episode or something like that." Gustav Söderström So that product couldn't have existed without generative AI, w summers ago," or "Here's some new stuff that were trending yesterday in The Last of Us episode or something like that." Gustav Söderström So that product couldn't have existed without generative AI, both generating the voice and generating the content of what the voice says. So you can have individualized, personalized voice at the scale of half a billion people. And so we had the use case we have seen for many, many years. Sometimes people call it the radio use case. We called it the zero intent use case internally when you actually don't know what you want to listen to at all. Gustav Söderström Spotify wasn't that good. Spotify was good when, at least roughly, you knew the use case of what you want to do, if it was a workout or dinner. We had lots of options for all of those. But if you really didn't know at all, it was hard to open Spotify and stare at it. And people used to say longingly that this was the one thing that radio was good at. Radio was quite bad, to be honest. I mean, it's not personalized to you at all. It's not on demand. You come in in the middle of things, it's actually terrible in many ways. But people still often say that there was something good about it. And I think that's something was the fact that you had a knob and you could just switch between contexts. It's like no, boring, boring, boring, boring, okay, this is good. Gustav Söderström And Spotify never had that mode of, I don't know what I want, but I want to cycle through things until I find something that I like. And I think with the AI DJ, that's actually the use case we managed to solve. So X comes on and says, "I'm going to suggest something to you that you can listen to." And if you like it, you can keep listening, but if you don't like it, you bring him back again and you change channel. And for one reason or another, we tried to solve that for many times for a long time, but just starting to play a random song without any context as to why you would hear this, it just never worked. So that was our first foray into a product that couldn't exist before. And I think to your question of principles around that, there are a few pretty distinct principles that we've learned. Gustav Söderström One that I really like that is not my principle at all, I think it is straight from Chris Dixon, is the principle of fault-tolerant user interfaces. So I can't say how many times during the early machine learning era when we said we're moving from curation to recommendation. I saw a design sketch that was a single big play button because clearly that is the simplest user interface you can do, but if you don't understand the performance of your machine learning, you can't design for it. The quality of your machine learning, if you're going to have a single play button, needs to be literally 100% or zero prediction error, and that's never the case. So let's say that you have a one in five hits, four out of five things are done, then you need a UI that probably at least shows five things at the same time on screen. So you have a one in five of something being relevant on screen.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.