The Exchanges

Q What with Suno did you not do that you wish you'd done?

A A very, a very obvious one for me is actually getting off of Discord. Um, uh, so we released our first product in, uh, August of, uh, of last year. Um, and in November, we put up a very thin web app, and I got this totally wrong. I said, we're gonna be on Discord forever. I look at Midjourney, they're printing money. It's just a Discord bot, and now it's not, but at the time it was. And I did not appreciate just how much a good UI will totally change the experience for people. Discord is not the best UI for what it is that we do. It's better for Midjourney than it is for music, but like so much more is possible. And so we released this app, we released this web app, which is not even the full functionality of the Discord bot in November, and it takes five days for 90% of the traffic to move over to the web. Five days. It's like, there's no world in which you can say that I got that right.

AI assessment note: “A very, a very obvious one for me is actually getting off of Discord.”

Q What with Suno did you not do that you wish you'd done?

A A very, a very obvious one for me is actually getting off of Discord. Um, uh, so we released our first product in, uh, August of, uh, of last year. Um, and in November, we put up a very thin web app, and I got this totally wrong. I said, we're gonna be on Discord forever. I look at Midjourney, they're printing money. It's just a Discord bot, and now it's not, but at the time it was. And I did not appreciate just how much a good UI will totally change the experience for people. Discord is not the best UI for what it is that we do. It's better for Midjourney than it is for music, but like so much more is possible. And so we released this app, we released this web app, which is not even the full functionality of the Discord bot in November, and it takes five days for 90% of the traffic to move over to the web. Five days. It's like, there's no world in which you can say that I got that right.

AI assessment note: “A very, a very obvious one for me is actually getting off of Discord.”

Q Dude, I, I already just love chatting to you, because we just go off in many different directions. Do you think scaling laws will continue? You said there about the generative improvements.

A For music, It's very different from text, and I think people will very sloppily look at the world of OpenAI and Anthropic and the hyperscalers and say, um, audio is just a couple years behind, which it is, but that scale is gonna solve all these things. But unlike those domains where you're trying to just get more and more answers to objective problems, like I wanna get a better SAT score, I wanna do better on this benchmark, music is totally subjective. And, uh, so scale is not the answer to all the problems, so the models stay relatively small, um, and there are other techniques that you have to use to actually have these things have good taste.

AI assessment note: “scale is not the answer to all the problems, so the models stay relatively small”

Q I love that as like a tagline, but I actually spoke to, uh, Mignano before the show, and he was like, dude, you gotta ask him about the origin, because it was an enterprise AI audio tool, and it wasn't what you see today. Can you talk to me about that pivot?

A I think, I, I wouldn't call it a pivot. You know, we always knew that Um, audio is really far behind the world of text. You know, that's what, where we came from. Uh, our backgrounds are all NLP, and we thought it would actually be a lot harder to do good generative stuff, and so we thought the first product would be, um, more sense-making. You know, try to, try to scratch your head and think back to, like, GPT-II. Um, no one was really making interesting text with GPT-II, but GPT-II was like this interesting tool for understanding text, and that's where we thought we would be stuck Uh, for a couple of years until we learned to scale these things up, and it turned out we were wrong, and that good generative capabilities came out much, much sooner, and so, um, we, we, we, I mean, the interesting thing here is the ability to generate stuff, and so we, we kind of very quickly threw out the sense making tool.

AI assessment note: “we kind of very quickly threw out the sense making tool”

Q Yeah, fuck that. Uh, I'm, I'm young, but not that young. Uh, why do physicists and economists make the, make the best machine learning engineers?

A Uh, great question. Two, two completely different reasons, I think, um, you know, and I'll preface this with like, uh, you know, building an AI company, uh, you are in the business of trying to find talent and trying to find underappreciated talent. Let's be honest, I will pay you less than OpenAI will pay you, and I need to find a reason to convince you to come join us. And I think, um, for economists, economists are, and I am not an economist, but are great at thinking about natural experiments. They're great about doing kind of first principles reasoning in a way that isn't just like turn the tank, turn the crank, get better at this benchmark, but it's much better at thinking about what do these benchmarks really mean? Are there natural experiments that I can pursue there? Because, um, lots of Um, economics research happens in, in kind of data-poor environments, and I think those are really interesting perspectives. Um, physics I'm maybe a little bit closer to. Experimental physicists just get good at running high-quality experiments really quickly, and, um, AI is an empirical discipline, and so whoever can run more high-quality experiments quickly will win.

AI assessment note: “Experimental physicists just get good at running high-quality experiments really quickly”

Q What do you mean by that? What did you learn from that valuable data?

A So, uh, imagine it were just, uh, only the free tier or everything were totally free. Um, when you hit that paywall, sorry, when you hit the end of your free tier, um, I don't know why. I don't know if you wanted to continue or didn't want to continue. Um, I don't know how to pick out the users that I want to interview that found this really valuable or not really valuable. Like, I'm kind of lost in the desert a little bit, and If you have it and you can say, okay, these are the people who subscribed. These are the people who subscribed before they even hit the paywall. Let's go talk to them. Let's figure out, like, what was that magical moment they had? Or these were the people that hit the paywall and didn't subscribe. Let's go talk to them. Let's figure out why we didn't wow them. These are the, and sure, maybe we would be a little bit bigger, but we would also have, I think, a worse product if we hadn't had that data at our hands.

AI assessment note: “we would also have, I think, a worse product if we hadn't had that data”

Q Has there ever been a time where revenue focus has led to deprioritization of growth?

A No, because we've always prioritized growth on top of revenues. You know, this is like, um, I, I, I think people, I think, I think good people have the capacity to keep both things in their head. It's like, yes, we want to grow, but we also don't want to sacrifice revenues. And if you, like, you know, hold my feet to the fire, maybe I can give you a, a quantitative number of, I care about growth twice as much as I care about revenues. But like, where we are now in our journey, uh, we are not really making trade-offs yet. You know, we've reached a tiny fraction of the addressable audience for the product as it is today. We are building much more product with way more valuable and fun experiences that will reach an even bigger audience. So like, if I am making trade-offs now, I'm doing something wrong.

AI assessment note: “No, because we've always prioritized growth on top of revenues.”

Q Yeah, I love that. Um, listen, final one. What question have I not asked you that I should have asked?

A I think you asked, and we didn't, and then we got sidetracked. It's like, what should the future of music be? And cause that's, yeah, like that's, that's, um, cause should is, is the real word there. Cause, cause we need to build it. And it's the question that I think about night and day more than, more than any other question. And I think the, the, the, the good outcomes here look like a lot more people doing a lot more music for a lot more hours of the day. And all of the economics will flow downstream from that. And I think about AI as a tool for achieving many of those ends. It lets a lot more people participate in creation. It lets us recommend music to people more. It lets us, um, it will let you tailor a song for your girlfriend to her tastes, which is like very, very powerful because it turns out she needs to like the song for that effect to, to happen. Um, it's one where a lot more people I think can make a living Doing music, and right now, not enough people can actually make a living doing music, and I think about Instagram, or I think about photography pre and post Instagram, very few people made a living doing photography, and now a lot more people can make a living doing photography because of Instagram, and yes, it changes the tenor of photography. Yes, it changes that every individual image is less valuable than it used to be, but in aggregate, this is a much, m…

AI assessment note: “It's like, what should the future of music be?”

Q So if we think about, like, insight developments, you said to me before, music should more closely resemble a video game in the future. I thought this was a fascinating analogy or comparison. Why should it resemble a video game?

A The thing about video games is that they're interactive, uh, they're engaging, they're rich experiences, they're, they're fun by yourself, they're more fun with your friends, and when I think about what music should be for me, uh, it should be all of those things. Um, and then, uh, so in that sense, I want to make music more like a video game. Nobody half plays video games the same way people kind of put on music in the background and half pay attention to it. Um, and then the other way is I think if you accomplish all those things and if you make music interactive and you make music engaging, um, people will pay for it like they pay for video games. And I'm sure I don't need to tell you that the video game industry is so much bigger than the music industry and, and most other industries. And it's because people have no problem parting with their hard earned money, um, to experience those things. And to me, it seems like just crazy that music should not be as engaging as Fortnite.

AI assessment note: “if you make music interactive and you make music engaging, um, people will pay”

Q What is the metric for a successful user? Is it three songs created, one song shared? How do you think about that?

A There's a few things we look at there. I think, um, the, the most salient one is actually just like, did you hit the paywall your first day? Um, because even if you didn't go through it for whatever reason, if you hit the paywall, you just enjoyed the last 10 or 12 minutes of your life, and I know I did something good there, and if you made one song and threw it away and never came back to it, we didn't wow you, you didn't have that magical experience, and we missed our mark. So I think that's, like, kind of the most important one, and it's, it's fairly high how the fraction of people that hit the paywall their first day.

AI assessment note: “the most salient one is actually just like, did you hit the paywall your first day?”

Q How do you compete in a war for talent with OpenAI, with Anthropic, with the massive, uh, incumbents that we have in the space who are paying millions of dollars?

A Yeah, we don't pay millions of dollars. Uh, so how do we compete? Um, there's a couple things, you know, one is, uh, we're not in Silicon Valley, um, and, and for a variety of reasons, but, um, if you want to be in one of the cities where we are, we're headquartered in Cambridge, Massachusetts. Um, we're kind of the, the coolest AI company, certainly in Cambridge, Massachusetts. Um, there's something else that goes back to what we were talking about before, which is that, um, if you are interested in a different AI problem, in the problem of how do I align Models, not to objective truth, but to human taste. I think there's nowhere else to do it.

AI assessment note: “if you are interested in a different AI problem... there's nowhere else to do it.”

Q It's pretty special. Um, tell me, what concerns you most in the world today?

A People are not good at realizing what is the first order effects of, of the issues they deal with or the things that they are doing. And people usually glom onto the second order effects because sometimes they, they like feel, I don't know if they feel smarter or it feels closer to home, but like people are not good at judging, at judging effect sizes. I'll give you, I'll give you some examples. Like, um, uh, I teach this course at, at, um, at Sloan at MIT's business school. So like you need to have an AI policy. Um, because it's a course. Like, are you going to allow students to use ChatGPT? And you get a lot of people saying, like, GPT is the end of education, you know. And my, my answer to that is, like, no. The first order effect here is that GPT means that every person in the world has, like, a median competent tutor or sidekick. This is obviously amazing for education, and if you can't recognize that fact, I'm not sure you have such good judgment. Now let's talk about the second order effect, which is that, like, Yeah, all my homeworks just got hackable by, you know, one good prompt, and my response to that is like, yeah, ok, so I, as the instructor, um, now need to change what I teach people, um, and that's scary, and that's a lot of work, but if I don't change what I teach people, I'm not really preparing them for the real world, right? Like, most companies are gonna le…

AI assessment note: “People are not good at realizing what is the first order effects”

Q Dude, I, I already just love chatting to you, because we just go off in many different directions. Do you think scaling laws will continue? You said there about the generative improvements.

A For music, It's very different from text, and I think people will very sloppily look at the world of OpenAI and Anthropic and the hyperscalers and say, um, audio is just a couple years behind, which it is, but that scale is gonna solve all these things. But unlike those domains where you're trying to just get more and more answers to objective problems, like I wanna get a better SAT score, I wanna do better on this benchmark, music is totally subjective. And, uh, so scale is not the answer to all the problems, so the models stay relatively small, um, and there are other techniques that you have to use to actually have these things have good taste.

AI assessment note: “scale is not the answer to all the problems, so the models stay relatively small”

Q Yeah, fuck that. Uh, I'm, I'm young, but not that young. Uh, why do physicists and economists make the, make the best machine learning engineers?

A Uh, great question. Two, two completely different reasons, I think, um, you know, and I'll preface this with like, uh, you know, building an AI company, uh, you are in the business of trying to find talent and trying to find underappreciated talent. Let's be honest, I will pay you less than OpenAI will pay you, and I need to find a reason to convince you to come join us. And I think, um, for economists, economists are, and I am not an economist, but are great at thinking about natural experiments. They're great about doing kind of first principles reasoning in a way that isn't just like turn the tank, turn the crank, get better at this benchmark, but it's much better at thinking about what do these benchmarks really mean? Are there natural experiments that I can pursue there? Because, um, lots of Um, economics research happens in, in kind of data-poor environments, and I think those are really interesting perspectives. Um, physics I'm maybe a little bit closer to. Experimental physicists just get good at running high-quality experiments really quickly, and, um, AI is an empirical discipline, and so whoever can run more high-quality experiments quickly will win.

AI assessment note: “economists are... great at thinking about natural experiments... Experimental physicists just get good at running”

Q What do you mean by that? What did you learn from that valuable data?

A So, uh, imagine it were just, uh, only the free tier or everything were totally free. Um, when you hit that paywall, sorry, when you hit the end of your free tier, um, I don't know why. I don't know if you wanted to continue or didn't want to continue. Um, I don't know how to pick out the users that I want to interview that found this really valuable or not really valuable. Like, I'm kind of lost in the desert a little bit, and If you have it and you can say, okay, these are the people who subscribed. These are the people who subscribed before they even hit the paywall. Let's go talk to them. Let's figure out, like, what was that magical moment they had? Or these were the people that hit the paywall and didn't subscribe. Let's go talk to them. Let's figure out why we didn't wow them. These are the, and sure, maybe we would be a little bit bigger, but we would also have, I think, a worse product if we hadn't had that data at our hands.

AI assessment note: “These are the people who subscribed... Let's figure out, like, what was that magical moment”

Q It's pretty special. Um, tell me, what concerns you most in the world today?

A People are not good at realizing what is the first order effects of, of the issues they deal with or the things that they are doing. And people usually glom onto the second order effects because sometimes they, they like feel, I don't know if they feel smarter or it feels closer to home, but like people are not good at judging, at judging effect sizes. I'll give you, I'll give you some examples. Like, um, uh, I teach this course at, at, um, at Sloan at MIT's business school. So like you need to have an AI policy. Um, because it's a course. Like, are you going to allow students to use ChatGPT? And you get a lot of people saying, like, GPT is the end of education, you know. And my, my answer to that is, like, no. The first order effect here is that GPT means that every person in the world has, like, a median competent tutor or sidekick. This is obviously amazing for education, and if you can't recognize that fact, I'm not sure you have such good judgment. Now let's talk about the second order effect, which is that, like, Yeah, all my homeworks just got hackable by, you know, one good prompt, and my response to that is like, yeah, ok, so I, as the instructor, um, now need to change what I teach people, um, and that's scary, and that's a lot of work, but if I don't change what I teach people, I'm not really preparing them for the real world, right? Like, most companies are gonna le…

AI assessment note: “People are not good at realizing what is the first order effects”

Q So if we think about, like, insight developments, you said to me before, music should more closely resemble a video game in the future. I thought this was a fascinating analogy or comparison. Why should it resemble a video game?

A The thing about video games is that they're interactive, uh, they're engaging, they're rich experiences, they're, they're fun by yourself, they're more fun with your friends, and when I think about what music should be for me, uh, it should be all of those things. Um, and then, uh, so in that sense, I want to make music more like a video game. Nobody half plays video games the same way people kind of put on music in the background and half pay attention to it. Um, and then the other way is I think if you accomplish all those things and if you make music interactive and you make music engaging, um, people will pay for it like they pay for video games. And I'm sure I don't need to tell you that the video game industry is so much bigger than the music industry and, and most other industries. And it's because people have no problem parting with their hard earned money, um, to experience those things. And to me, it seems like just crazy that music should not be as engaging as Fortnite.

AI assessment note: “they're interactive, uh, they're engaging... music should be... all of those things.”

Q we like really dive into the company. There were elements that we were talking about before where you were like, I'd like to touch on these and they were not in my purview of like, okay, these are logical ones to go with. You said to me, quantum computing will be amazing But you still should not do it. Why will it be amazing, but why shouldn't we do it?

A I think the, the promises are, are, are incredible. You know, there's a big part of me that wishes that we knew more things to do with these quantum computers, but I just see a rush of, um, commercial dollars into the space, um, where far more research was necessary, and people are so short-sighted, including, if I may, in your business, VCs are so Short-sighted. You expect returns over a certain time horizon, and at least from my vantage point, the, the challenges are not, um, they're still physics challenges. They're still like basic research challenges that have to happen here, and so it's not something where you can kind of just turn the crank and make companies out of it, and to my knowledge, all of the companies that sprang up around this are not doing all that well, and to my knowledge, the best stuff that happens, happens in this weird public-private Partnership, uh, type of situation.

AI assessment note: “the challenges are not, um, they're still physics challenges. They're still like basic research challenges”

Q How do you compete in a war for talent with OpenAI, with Anthropic, with the massive, uh, incumbents that we have in the space who are paying millions of dollars?

A Yeah, we don't pay millions of dollars. Uh, so how do we compete? Um, there's a couple things, you know, one is, uh, we're not in Silicon Valley, um, and, and for a variety of reasons, but, um, if you want to be in one of the cities where we are, we're headquartered in Cambridge, Massachusetts. Um, we're kind of the, the coolest AI company, certainly in Cambridge, Massachusetts. Um, there's something else that goes back to what we were talking about before, which is that, um, if you are interested in a different AI problem, in the problem of how do I align Models, not to objective truth, but to human taste. I think there's nowhere else to do it.

AI assessment note: “we're headquartered in Cambridge, Massachusetts... interested in a different AI problem”

Q What is the metric for a successful user? Is it three songs created, one song shared? How do you think about that?

A There's a few things we look at there. I think, um, the, the most salient one is actually just like, did you hit the paywall your first day? Um, because even if you didn't go through it for whatever reason, if you hit the paywall, you just enjoyed the last 10 or 12 minutes of your life, and I know I did something good there, and if you made one song and threw it away and never came back to it, we didn't wow you, you didn't have that magical experience, and we missed our mark. So I think that's, like, kind of the most important one, and it's, it's fairly high how the fraction of people that hit the paywall their first day.

AI assessment note: “the most salient one is actually just like, did you hit the paywall your first day?”

Q Has there ever been a time where revenue focus has led to deprioritization of growth?

A No, because we've always prioritized growth on top of revenues. You know, this is like, um, I, I, I think people, I think, I think good people have the capacity to keep both things in their head. It's like, yes, we want to grow, but we also don't want to sacrifice revenues. And if you, like, you know, hold my feet to the fire, maybe I can give you a, a quantitative number of, I care about growth twice as much as I care about revenues. But like, where we are now in our journey, uh, we are not really making trade-offs yet. You know, we've reached a tiny fraction of the addressable audience for the product as it is today. We are building much more product with way more valuable and fun experiences that will reach an even bigger audience. So like, if I am making trade-offs now, I'm doing something wrong.

AI assessment note: “No, because we've always prioritized growth on top of revenues.”

Q we like really dive into the company. There were elements that we were talking about before where you were like, I'd like to touch on these and they were not in my purview of like, okay, these are logical ones to go with. You said to me, quantum computing will be amazing But you still should not do it. Why will it be amazing, but why shouldn't we do it?

A I think the, the promises are, are, are incredible. You know, there's a big part of me that wishes that we knew more things to do with these quantum computers, but I just see a rush of, um, commercial dollars into the space, um, where far more research was necessary, and people are so short-sighted, including, if I may, in your business, VCs are so Short-sighted. You expect returns over a certain time horizon, and at least from my vantage point, the, the challenges are not, um, they're still physics challenges. They're still like basic research challenges that have to happen here, and so it's not something where you can kind of just turn the crank and make companies out of it, and to my knowledge, all of the companies that sprang up around this are not doing all that well, and to my knowledge, the best stuff that happens, happens in this weird public-private Partnership, uh, type of situation.

AI assessment note: “the challenges are... still physics challenges. They're still like basic research challenges”

Q How do you align models to human taste when human taste is so subjective?

A Yeah, it's really hard. Uh, you know, and I think, um, we have the advantage of having a lot of usage, and so we can run, um, we have a lot of data, we can collect a lot of data, we can run a lot of A-B tests on things. Um, in the future, there's probably a lot more personalization that happens in this domain, but in the meantime, uh, it's similar techniques that align models to, Uh, human preferences for like RLH effort for, um, chat GPT, et cetera. Um, but it's totally not obvious that that is what the future should be. You know, it's totally not obvious that the same techniques that are used to align, um, LLMs to weird human taste should be the same techniques that we use to align music models.

AI assessment note: “it's similar techniques that align models to, Uh, human preferences for like RLH effort”

Q When we think about the future of music then, how does that look? Because you have existing players today. Your Spotify is most namely of the world. How does that look in your ideal world?

A There's a few different ways. I'll tell you what I hope it is, which is that, um, there are a lot more people participating, and there are a lot more experiences on tap. And so what does that mean? That means, um, we didn't just want to build, let's say, a company that makes, uh, the current crop of creators 10% faster, or makes it 10% easier to make music. Um, if you want to impact the way a billion people experience music, you have to build something for a billion people. And so, um, that is first and foremost giving everybody the joys of creating music. And this is a huge departure from how it is now. It's not really enjoyable to make music now. People enjoy-

AI assessment note: “I'll tell you what I hope it is, which is that, um, there are a lot more people participating”

Q How do you align models to human taste when human taste is so subjective?

A Yeah, it's really hard. Uh, you know, and I think, um, we have the advantage of having a lot of usage, and so we can run, um, we have a lot of data, we can collect a lot of data, we can run a lot of A-B tests on things. Um, in the future, there's probably a lot more personalization that happens in this domain, but in the meantime, uh, it's similar techniques that align models to, Uh, human preferences for like RLH effort for, um, chat GPT, et cetera. Um, but it's totally not obvious that that is what the future should be. You know, it's totally not obvious that the same techniques that are used to align, um, LLMs to weird human taste should be the same techniques that we use to align music models.

AI assessment note: “it's similar techniques that align models to, Uh, human preferences for like RLH effort”

Q What, what are some of those business models then?

A Uh, imagine you found some creator that you really love, uh, who is a teenager in Saskatchewan, and you just love this guy's music, and, um, you could have a Patreon thing where you can pay him directly. You could have, uh, your own, um, kind of fork of his model if, uh, if he were cool with that and you were cool with that, and you're paying him for that, and now you're making songs that are, you know, your riffs on what he does. I think Let me give you an example. Um, we had this remix contest with Timbaland, and, um, a tremendous number of people, uh, submitted remix songs, and to me, getting to remix the music of your musical idol is, like, the ultimate form of, um, of engagement with them. It is so much cooler than, like, honestly, even meeting them backstage after a concert, right? Like, the old model of StreamShare does not Properly account for an interaction like that.

AI assessment note: “You could have, uh, your own, um, kind of fork of his model”

Q And, um, I'm quite energized by the fact that, like, I think we have- How can you as a public company CEO, say of a UMG, How can you say we know it's coming and not address the core of creation?

A I mean, I think what that person would do in that case is say, yes, AI is an important part of the creation stack right now, um, and not try to address the, like, well, what about all the other people who, who want to create with these new tools also? Um, I think, I think, but, um, I, I think it, it applies just as much as, uh, to me as it does to the, To the CEO of the Universal Music Group of like, we can actively build a bigger, better future of music with AI, or we can just wait, and someone else in another country, not bound by US laws, not with the same intentions, um, will build a worse future of music with AI. And I can think of lots of dystopian futures. Uh, just to name like two particularly bad ones. One would be like a group Uh, in another country that doesn't want to follow the laws, we'll make it so that you can, without permission, just impersonate your favorite artist and just make endless copies, you know, endless Ariana Grande songs, like you said, um, without giving a cent to Ariana Grande. Like, that would not be great. And-

AI assessment note: “we can actively build a bigger, better future of music with AI”

Q What, what are some of those business models then?

A Uh, imagine you found some creator that you really love, uh, who is a teenager in Saskatchewan, and you just love this guy's music, and, um, you could have a Patreon thing where you can pay him directly. You could have, uh, your own, um, kind of fork of his model if, uh, if he were cool with that and you were cool with that, and you're paying him for that, and now you're making songs that are, you know, your riffs on what he does. I think Let me give you an example. Um, we had this remix contest with Timbaland, and, um, a tremendous number of people, uh, submitted remix songs, and to me, getting to remix the music of your musical idol is, like, the ultimate form of, um, of engagement with them. It is so much cooler than, like, honestly, even meeting them backstage after a concert, right? Like, the old model of StreamShare does not Properly account for an interaction like that.

AI assessment note: “your own, um, kind of fork of his model if, uh, if he were cool”

Q When we think about the future of music then, how does that look? Because you have existing players today. Your Spotify is most namely of the world. How does that look in your ideal world?

A There's a few different ways. I'll tell you what I hope it is, which is that, um, there are a lot more people participating, and there are a lot more experiences on tap. And so what does that mean? That means, um, we didn't just want to build, let's say, a company that makes, uh, the current crop of creators 10% faster, or makes it 10% easier to make music. Um, if you want to impact the way a billion people experience music, you have to build something for a billion people. And so, um, that is first and foremost giving everybody the joys of creating music. And this is a huge departure from how it is now. It's not really enjoyable to make music now. People enjoy-

AI assessment note: “I'll tell you what I hope it is, which is that”

Q Um, now, I'm just starting with a blunt question. There is lawsuit from RIAA in June, 2024, um, whatever stuff. Was there any basis to this lawsuit? The lawsuit is basically saying that you use their media to train your models.

A I have to be very careful what I say and don't say about a lawsuit. So, um, I think I'll tell you, yeah, we know that there are some copyrighted works in our training data. Um, that's not illegal. Um, it's stock standard for the industry. It's what every AI company does. Um, You know, I think in some sense the lawsuit wasn't totally shocking. You know, it's, um, most AI companies get sued. Everybody in music gets sued. It's a highly litigious industry. I, I think it's, it's a little bit, um, it's a little bit depressing in some, in, in some way because I think there is a much bigger and brighter future of music to build together with the existing industry instead of, ah, kind of fighting it out and having the potential for this thing to Um, just be net smaller. You know, I, I'll tell you, um, I'll say something ill of, of, of lawyers for a minute, uh, but we were talking about economists, and, um, uh, there's this famous, um, econ paper from, like, the eighties, uh, looking at what, why do, why do some countries grow and some countries don't grow, and, and, um, by this guy Andre Schleifer, and, um, one of the conclusions is basically, like, they're looking at the ratio of, like, how many engineers are in a country and how many lawyers are in a country, And the, the conclusion is like, more engineers equals more growth, more lawyers equals less growth. And, you know, obviously t…

AI assessment note: “yeah, we know that there are some copyrighted works in our training data”