Mikey Shulman, CEO @Suno: The Future of Music, What is Gonna Happen? | E1244

OpenAI is amazing. They did every AI company a huge disservice because everybody thinks that just, like, empty textbox is now the right interface. And it is for ChatGPT and it is incorrect for basically everything else. At some point, I don't know if it's version four or version five, there will be a last model release that is released as a model, and everything else is just product releases.

Ready to go? (upbeat music plays) Mikey, I am, like, the biggest fanboy of Suno, and I have created like 25 songs in the last few days. So, I'm so excited for this. Thank you for joining me.

So good to be here. Thank you for having me.

Now, I would love to just start with, for those that don't know what Suno is, what is Suno and what is Suno not?

Uh, yeah, Suno is a way, um, for everybody to experience, uh, all of the joys of music, meaning not just background listening to music, but, uh, losing yourself in the process of making music, sharing music, of editing music, um, of being a much more active participant in music. You know, I like to think we're not making music, we're making musicians.

I love that as, like, a tagline. But I actually spoke to, uh, Minjano 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?

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 wh- 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-2. Um, no one was really making interesting text with GPT-2, but GPT-2 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... The- the interesting thing here is the ability to generate stuff, and so we- we kind of very quickly threw out the sense-making tool.

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

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.