Mikey Shulman，Suno CEO：音乐的未来，会发生什么？| E1244

OpenAI is amazing. They did every AI company a huge disservice because everybody thinks that an empty text box is now the right interface. It is for ChatGPT, and it is incorrect for basically everything else.

Mikey, I am the biggest fanboy of Suno. I have created about 25 songs in the last few days, so I am very excited for this. Thank you for joining me.

1. What is Suno?

It is so good to be here. Thank you for having me.

I would love to start with what Suno is, for those who do not know. What is Suno, and what is Suno not?

Suno is a way for everybody to experience all of the joys of music. That means not just background listening to music, but losing yourself in the process of making music, sharing music, and editing music. It means being a much more active participant in music.

I like to think that we are not making music; we are making musicians. I love that as a tagline.

I spoke to Mina before the show, and he said, “You have to ask him about the origin,” because Suno was an enterprise AI audio tool and was not what we see today. Can you talk to me about that pivot?

I would not call it a pivot. We always knew that audio was really far behind the world of text. That is where we came from; all of our backgrounds are in NLP.

We thought it would actually be a lot harder to do good generative work, so we thought the first product would be more about sense-making. Try to think back to GPT-2: no one was really making interesting text with GPT-2, but GPT-2 was an interesting tool for understanding text.

That is where we thought we would be stuck for a couple of years, until we learned to scale these things up. It turned out we were wrong, and good generative capabilities came out much, much sooner.

The interesting thing here is the ability to generate stuff, so we very quickly threw out the sense-making tool.

I already love chatting to you because we go off in many different directions. Do you think scaling laws will continue?

You said that generative improvements for music are very different from text. I think people will very sloppily look at OpenAI, Anthropic, and the hyperscalers and say, “Audio is just a couple of years behind,” which it is, but that scale is going to solve all these things.

Unlike those domains, where you are trying to get more and more answers to objective problems—“I want to get a better SAT score” or “I want to do better on this benchmark”—music is totally subjective. Scale is not the answer to all of the problems.

2. Why Should Music Resemble Video Games?

The models stay relatively small, and there are other techniques that you have to use to actually have these things exhibit good taste. We live in a world of many small models in music. I do not know about many, but scale is not as much of a panacea as it is in text.

If we think about insightful developments, you said to me before that music should more closely resemble a video game in the future. I thought that was a fascinating analogy. Why should it resemble a video game?

3. What Does Mikey Regret Not Doing?

The thing about video games is that they are interactive, engaging, and rich experiences. They are fun by yourself, and they are more fun with your friends. When I think about what music should be for me, it should be all of those things.

In that sense, I want to make music more like a video game. Nobody half-plays video games in the same way that people put on music in the background and half-pay attention to it.

The other way is that if you accomplish all those things—if you make music interactive and engaging—people will pay for it like they pay for video games. I am sure I do not need to tell you that the video game industry is so much bigger than the music industry and most other industries.

It is because people have no problem parting with their hard-earned money to experience those things. To me, it seems crazy that music should not be as engaging as Fortnite.

4. Are Higher Prices for Better Models Just Temporary?

I agree with you that people will pay for it. I did the $300 option as fast as I could.

I tweeted this yesterday because of this, but I do not think that in the future we will pay more for access to newer models. I think that is just a stage of the development cycle that we are in. In 5 years’ time, we will not pay more for a better model; we will just get the product.

I agree with that very strongly. I am not even sure you will know what model there is.

At some point—I do not know if it is version 4 or version 5—there will be a last model release that is released as a model, and everything else will just be product releases. In the end, it is not today, but people are not going to care what powered the thing. You are just going to care that the music made you feel a certain way, and there are going to be lots of things that go into that.

Can I ask what powers Suno?

You mean the model?

Yes.

It is a Transformer model. We have not been shy about that.

Our competitive edge here has not been to innovate on the architecture. It has been to innovate on the audio representation. It is not obvious how to tokenize audio, but if you spend night and day doing that and basically take everything from the open-source text community about how to make and scale models, that is a pretty good recipe.

I want to go back to something you said: “We are not cheapening music; we are making it more valuable.” With total respect, as an economist—which I am not—but as a venture investor, I aim to be something I am not as a frequent job, when you increase supply, it reduces the price.

You are making supply practically infinite. Does that not reduce the price?

I think it reduces the average value of any given piece of content, but I think it greatly increases the value of music to society or to any given person.

That is what we want, right? We want a bigger, more vibrant music industry that has many more participants and much more engagement. We do not want this more precious thing where very few people have access to actually doing it and making it, and where the range of experiences that the average person has is limited and therefore not valuable.

5. Why Is Quantum Computing Amazing but Not Worth Pursuing?

Before we really dive into the company, there were elements that we talked about where you said, “I would like to touch on these,” and they were not in my purview of the logical questions to ask.

You said to me, “Quantum computing will be amazing, but you still should not do it.” I asked why it will be amazing, but why should we not do it?

The promises are incredible. There is a big part of me that wishes we knew more things to do with these quantum computers, but I see a rush of commercial dollars into the space where far more research was necessary.

People are so shortsighted, including, if I may, in your business. VCs are so shortsighted. You expect returns over a certain time horizon, and at least from my vantage point, the challenges are not solved. There are still physics challenges and basic research challenges that have to happen here.

It is not something where you can just turn the crank and make companies out of it. To my knowledge, all of the companies that sprang up around this are not doing all that well.

To my knowledge, the best work happens in this weird public-private partnership-type situation.

If you agree that it is amazing, should we not pursue it? Should we not go through that trough of disillusionment and endure the pain to get to the amazing outcome?

We should, but as career advice, people should not do it. It is a very risky career because it may not work.

People should not join companies doing it just yet. To be spicy, people should not join companies doing it just yet.

But if that is the case, we will never get to that end.

I agree with you. It is like a tragedy of the commons, but it is an irrational decision right now.

It is going to need some massive government intervention, I think in a good way, to make it happen. We should do that.

Should the government fund it?

I think so. Not venture investors.

I do not think you are going to get your money back in a reasonable time.

This is decades.

Call it enterprise SaaS investing, my friend.

Do you know how long it takes?

Yes, years.

No, but this is decades.

This is my point.

I am young, but not that young.

6. Why Do Physicists and Economists Excel in Machine Learning?

Why do physicists and economists make the best machine learning engineers?

Great question. There are 2 completely different reasons.

Building an AI company means trying to find talent and trying to find underappreciated talent. Let us be honest: I will pay you less than OpenAI will pay you, and I need to find a reason to convince you to join us.

For economists—and I am not an economist—they are great at thinking about natural experiments. They are great at doing first-principles reasoning in a way that is not just, “Turn the crank and get better at this benchmark.” They are much better at thinking about what these benchmarks really mean and whether there are natural experiments that they can pursue.

A lot of economics research happens in data-poor environments, and I think those are really interesting perspectives.

7. How Do You Compete for Talent Against AI Giants?

Physicists, particularly experimental physicists, get good at running high-quality experiments really quickly. AI is an empirical discipline, and whoever can run more high-quality experiments quickly will win.

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

We do not pay millions of dollars.

There are a couple of things. One is that we are not in Silicon Valley. For a variety of reasons, if you want to be in one of the cities where we are headquartered—in Cambridge, Massachusetts—we are the coolest AI company, certainly in Cambridge, Massachusetts.

There is something else that goes back to what we were talking about before. If you are interested in a different AI problem—the problem of how to align models not to objective truth but to human taste—I think there is nowhere else to do it.

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

It is really hard. We have the advantage of having a lot of usage, so we can run a lot of experiments. We have a lot of data, we can collect a lot of data, and we can run a lot of A/B tests on things.

In the future, there will probably be a lot more personalization than happens in this domain. In the meantime, the techniques are similar to those that align models to human preferences, such as RLHF for ChatGPT.

It is totally not obvious that this is what the future should be. It is totally not obvious that the same techniques used to align LLMs to weird human tastes should be the techniques we use to align music models.

How far will large model providers go into the application layer? I am really throwing you off here, but there is nothing to say that OpenAI does not move into customer support or music creation.

Absolutely. We think about this a lot. On the other hand, we kind of have to ignore it because we are just doing this.

I think they will try. I do not know if it will be music. Customer support seems honestly bigger. The thing that OpenAI is going after is—let us be honest—I hope to build a giant company, but general intelligence and customer support are both much bigger than music.

Customer support is not bigger. Zendesk is a $10 billion company at best, and it is being overbid by providers.

The future of customer support might be much, much bigger than Zendesk. If everything can be automated, you have all these BPOs. It is not obvious to me.

I am quite certain that the future of music is also much bigger than it is today, but maybe that has not dawned on the folks at OpenAI.

What I will say is that this is not just about scale. It is a product game at the end of the day. People will use Suno because it is a better product, not because there is a better model under the hood.

We are in the business of selling pleasurable musical experiences to people, plain and simple. A chat interface is not the be-all and end-all of what people want to do.

What is amazing about your product is your time to wow, which is about 5 seconds. I typed in, “Slow inspirational song to run to, male guitar vocals,” and it came out with something unbelievable in 5 seconds.

I am very happy we use Transformers because it lets us get there more quickly.

We can measure it. What you said is super important, and we keep a very close eye on making sure we deliver an amazing experience for your first song because it matters a ton.

For example, we can introduce a little bit of artificial latency into the product. We can see that people like it less. Sometimes you see the product working in the background, loading some magic, and people feel like more value is being created by the psychological experiment of making it take a little more time.

It is fascinating.

Ten seconds is worse than 8. I do not see us getting to 1 second anytime soon, but I think the crossover you are talking about is still far away.

8. What Defines a Successful User in Suno?

What is the metric for a successful user? Is it 3 songs created? Is it 1 song shared? How do you think about that?

There are a few things we look at. The most salient one is actually, “Did you hit the paywall on your first day?”

Even if you did not go through it for whatever reason, if you hit the paywall, you enjoyed the last 10 or 12 minutes of your life. I know I did something good there.

If you made 1 song, threw it away, and never came back to it, we did not wow you. You did not have that magical experience, and we missed our mark.

That is the most important one. The fraction of people who hit the paywall on their first day is fairly high.

This may be completely useless user feedback, but I found the numbers jarring because you price it annually. When you see $300, you think, “Wow, that is a lot.” If you showed me $29.99, I would think, “That is fair enough.”

We do want you to do an annual subscription, though.

It was not the most deliberate decision, but we decided to start charging from day 1, which is a little bit against the traditional wisdom of Silicon Valley: give the product away for free, scale, scale, scale.

We did not want to be a novelty item. We wanted to give people something they enjoyed enough to pay for.

When we launched, it was a Discord bot. We started collecting subscribers, and we always had the same free tier that we have today. If you wanted to do more, you had to pay.

A surprising number of people subscribed in that first month, and it was a sign that this was a real thing. Even without the revenues—which are fantastic to have and offset a lot of GPU burn—knowing what gets people to subscribe is such valuable data that I do not know how we would do this without it.

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

Imagine it were only the free tier, or everything were totally free. When you hit the end of your free tier, I would not know why. I would not know if you wanted to continue or did not want to continue.

I would not know how to pick out the users I want to interview—those who found this really valuable or those who did not find it valuable. I would be lost in the desert.

If you have the data and can say, “These are the people who subscribed before they even hit the paywall; let us talk to them and figure out what magical moment they had,” or, “These were the people who hit the paywall and did not subscribe; let us talk to them and figure out why we did not wow them,” that is incredibly useful.

Maybe we would be a little bit bigger without that data, but we would also have a worse product.

You mentioned GPU burn and revenue offsetting that. What percentage of your spend today is GPU burn, off the top of your head?

It is the biggest thing by far.

Really?

It is a few times payroll.

Wow.

We have a large research cluster as well. These technologies did not exist when we started the company, and research is still a huge part of what we do.

How do you anticipate that changing going forward?

I hope GPU prices come down, but we will probably just get more.

So you just get more, and the product gets better and better?

No. You need to run high-quality experiments.

I think the machine learning team at Suno will almost certainly not scale as quickly as a software team. Said differently, research is not something that you can solve with scale. You cannot just throw more bodies at it and get comparatively more results. There will be sublinear returns.

It is very important to understand that you have really talented researchers working on the problems that are going to be really important for the future of music.

I want to discuss the future of music and go to the product itself. I love the output, by the way.

I will start with a blunt question. There was a lawsuit from likely RIAA in June 2024. Was there any basis to that lawsuit? The lawsuit essentially says that you used their media to train your models.

I have to be very careful about what I say and do not say about a lawsuit.

I will tell you that we know there are some copyrighted works in our training data. That is not illegal. It is standard for the industry; it is what every AI company does.

The lawsuit was not totally shocking. Most AI companies get sued, and everybody in music gets sued. It is a highly litigious industry.

I think it is a little bit depressing because there is a much bigger and brighter future of music to build together with the existing industry, instead of fighting it out and having the potential for this thing to be net smaller.

I will say something ill of lawyers for a minute. We were talking about economists, and there is this famous economics paper from the 1980s looking at why some countries grow and some countries do not grow. It was by likely Andrei Shleifer.

One of the conclusions was basically that they looked at the ratio of engineers to lawyers in a country. More engineers equals more growth; more lawyers equals less growth.

Obviously, that is reductive, but instead of fighting it out, if we were working together to build a bigger future of music, everybody would be happier.

The music business has such an embedded fixed-pie mentality. There is a fixed pie of money out there, and we are all just trying to divide it unfairly for ourselves. If we focused on growing that pie, everything would get easier.

9. How Will AI Startups and Incumbents Resolve Conflicts?

You mentioned that it is natural for AI companies, whether OpenAI or any of the others, to face lawsuits and the same pressures. When you think about how this naturally gets resolved between new AI players and traditional incumbents in any industry—not specifically to this one—how do you think that plays out?

Is it settlements, with new AI companies bluntly using venture dollars to pay off large incumbents? Is it equity distributions to larger incumbents so they feel like they have a part of the companies?

I do not know. There is the traditional Silicon Valley mentality of, “Screw you, existing industry. I will disrupt you, and there is nothing you can do about it.” Then there is the existing-incumbent approach of, “We will sue you until you go away.”

Both of those are obviously wrong.

People will criticize us for building a tech company in Cambridge, Massachusetts, but it is actually nice not to have that embedded Silicon Valley mentality. Every single person at Suno has an incredible, deep love and respect for music.

Culturally, that is amazing, but it also keeps people out of that mentality and makes them much more focused on building a bigger, brighter future, hopefully with the incumbents in the industry.

There are plenty of incumbents in the industry who do work with us.

How would you have liked the incumbents who are suing you to have behaved?

Talk to us first, probably.

Some did, and some did not. On paper, it seems silly to throw a bunch of venture dollars at lawyers instead of sitting down and talking about how you could work together.

Deciding to sue first and ask questions later seems inefficient, but what do I know?

Does it worry you as a founder when you get a lawsuit? How do you respond?

Of course. I would obviously rather not get sued.

I know some genuinely psychotic founders who say, “Fuck you. Feed off the pain. This means I am winning. You notice me because you are suing me.” There is certainly an element of that, but those people are lying if they tell you this is a good thing.

Maybe it is a sign of a good thing, but it is not itself a good thing. If nothing else, it is going to cost money and time.

What happens if this lawsuit goes to trial and you lose?

It is obviously not good for us. The company is not dead, but it is obviously not good for us.

I do not think that is likely. I would be thinking about the game theory of what all players want here and what the best future of music is for everyone.

Even if you could make Suno-like companies or AI companies go away, do you really want that? What if the music industry could be as big as the gaming industry? There would be a lot of happy people.

10. How the Future of Music Looks Like?

When we think about the future of music, how does that look? You have existing players today, most notably Spotify. How does that look in your ideal world?

There are a few different ways. I will tell you what I hope it is: there are many more people participating and many more experiences on tap.

What does that mean? We did not just want to build a company that makes the current crop of creators 10% faster or makes it 10% easier to make music.

If you want to impact the way 1 billion people experience music, you have to build something for 1 billion people. First and foremost, that means giving everybody the joys of creating music.

That is a huge departure from how it is now. It is not really enjoyable to make music now. People say it takes a lot of time and practice, and you need to get really good at an instrument or a piece of production software.

I think the majority of people do not enjoy the majority of the time they spend making music.

Do not think that is like running. Running is hard and painful. You do not particularly enjoy it, but you love running, you get good at it, and you get better at it. You speak to runners, and they love running.

Most people drop out of that pursuit because it was hard. The people you know who run are a highly biased selection of the population that fell in love with it.

I am jumping back and forth, but we talked about training and you said you train on copyrighted music. At what point can you train on Suno music, because you have such a library now? When can AI train on AI?

That is a huge, open question. How do you do this in a way that does not impart massive problems into models?

We work on it a little bit. I think the bigger players are hitting data walls and are more actively working on it. Like everything else, it is really nice to have the big players solve a lot of the thorny problems for you so you can focus on your competitive advantage.

I am hoping somebody solves that problem and it is not us.

When we spoke before, you said it is not good to have 2 worlds—AI music and normal music. What does that look like from a customer experience?

Why is Spotify not simply blended into the app, so you have both in one home?

From a customer perspective, at the very least, it would be a bummer if I had to go to 2 different places.

You use Spotify. I do not know if there is any content you know about that does not exist on Spotify that you are looking for, but it is a huge pain in the butt to go to another app to find it.

There are some audiobooks that are not there yet. I just say, “Fuck it, I will wait.”

That shows you there is a lot of friction.

If there were 2 worlds of music, that would be friction for the average consumer. But I actually hope it is much more than just consumption.

There should not be a set of fun experiences around making and sharing music that is disjointed from all the other music you listen to at a party, while you work out, in your car, or while commuting.

I think these things together are much bigger than if you separate them.

I totally agree with the separation and the friction that comes with that, but that means either you need to move into Spotify or Spotify needs to move into you.

Or both, or neither. There are lots of different worlds that are possible here. Time will tell which of these things is correct.

How do you think artists feel about Suno?

That is a great question. I will tell you that, obviously, they love it.

Actually, it is a biased sample of artists that I speak to, but the vast majority of artists I speak to behind closed doors admit that they use and love Suno. There are only a couple I have ever met who are not into it.

11. Can Artists Have Personalized AI Models for Their Music?

This makes me extremely optimistic about the future.

Is there a way to create personalized models for different creators? What I mean is, Ariana Grande comes to you and says, “I want to do a deal with you where I give you exclusive access to my content. You train on my content, I leverage my brand, and I effectively have full control but unlimited supply of future Ariana Grande songs in seconds.”

The answer is yes, and I would love to do that.

Depending on what her contracts look like, she might not even be allowed to do that, which would be crazy. Why would you not be able to?

She does not own all of her music, but she owns her name and likeness. I would love to get to a world where she can have models that make Ariana Grande songs.

You have noticed that in our product you cannot make Ariana Grande songs. If you put her name in, we will wag our finger at you, and you will say, “That is not what the future of music is.”

Suno is for original music. It is not for impersonating people.

But if she uses it, it is not impersonating her, right?

Exactly. This is an immensely powerful tool for her to make music for people writing songs for her, to hear how she would do it, or maybe even for her to give it to her superfans.

Her superfans could make what I can only describe as the equivalent of fan fiction, if she wants that. In my mind, that would be an immensely engaging activity—way more valuable and engaging than having an AMA with her if you are a huge fan.

I love Dean Lewis. I love this kind of male guitarist using Suno songs. Clearly, I need a therapist.

If I click “Go to Radio” on Spotify, I hear other songs like his. He does not get paid for other songs like his.

What is the difference between me going to Suno and saying, “Give me some songs like Dean Lewis, but focused on an Australian man and an Australian woman, sadly sung with guitars,” and him getting a cut because Dean Lewis is in the prompt?

Right now, you cannot do that. I would love to see a future where you can, with him opting into that as an example. How does that completely change what you can do?

The thing that people do not realize is that right now we have made something where people enjoy making music so much that they are paying to create the music. This has nothing to do with listening.

I think a big part of the future of music is rethinking a lot of these business models. Right now, it is set up around a capped stream share. The stream share is a certain size, everybody is fighting over it, and artists do not make a ton of money from it.

Hopefully, there are new business models that are much more tied to the enjoyment people get and are much more tied to artists.

If people want to interact with the artist themselves, what are some of those business models?

Imagine you found a creator who you really love, who is a teenager in Saskatchewan, and you just love this guy’s music.

You could have a Patreon where you pay him directly. You could have your own fork of his model if he were comfortable with that and you were comfortable with that, and you would pay him for it.

Now you are making songs that are your riffs on what he does.

Let me give you an example. We had this remix contest with Timbaland (likely), and a tremendous number of people submitted remixed songs.

To me, getting to remix the music of your musical idol is the ultimate form of engagement with them. It is so much cooler than, honestly, even meeting them backstage after a concert.

The old model of stream share does not properly account for an interaction like that.

Talk to me about that partnership. Timbaland (likely) is obviously a huge industry figure. How does that even come about?

If you talk to anyone in the business, they will tell you that he is at the cutting edge of technology and music, and always has been.

He was a fan of the product before we met him, which is amazing. You cannot fake this stuff.

It is an amazing partnership for a variety of reasons. We learned a ton working with him. We do a lot of product development with him, and he has access to things before anyone else does.

The next thing is that this provides some amount of recognition for us.

Do you think it legitimizes the traditional music industry?

Yes, I do.

The real thing this does is that you have someone who has made it and does not need to do this publicly saying, “Yes, I use Suno, and it is awesome.”

That gives cover to up-and-comers to be more open and vocal about using it.

Can I be blunt? Do you pay him for that partnership?

No, we do not pay him in cash.

Does he get equity?

He is an adviser to the company.

Fantastic. I think Calm did this right.

I do not know. Calm did it phenomenally well, and with some incredible people.

Would you like to do more of these partnerships? When you think forward about distribution and legitimacy, do you think that is important?

Totally. We always wanted to do this, and we continue to want to do this with players in the existing industry.

Timbaland (likely) was not the first, and he certainly will not be the last.

A lot of pop music has gotten boring. Why has pop music gotten boring?

We could talk about this for hours.

I think songs have gotten shorter, but also much more homogeneous in melodies, harmonies, and song structures. Music innovates very much sonically. There are really interesting sounds, but they all fit into the same song structure.

This is a product of the technology people use to make music, digital production, streaming, platforms like TikTok—it is everything.

We have seen this with Olivia Rodrigo. People are now changing the structure of songs to align with the algorithms of social channels.

Absolutely.

How do you think about that? Is it sad?

I do not think it is obviously good or bad. It just is.

Those algorithms have spread that music to many more people than otherwise would have found it.

12. How Will Music Discovery Evolve with Infinite Supply?

You mentioned discovery when you talked about making more music. How do you think about the future of music discovery when there is an infinite supply of music?

In some sense, it gets harder. In some sense, it gets easier.

People do not realize the extent to which discovery is already extremely algorithmic, and that you are not always being fed as much of the long tail as you might want—or maybe it is exactly what you want.

What do people think about music discovery that they get wrong?

I think people do not realize the extent to which the music that is popular is a product of recommendation algorithms.

You should ask somebody at Spotify or TikTok, not me. I think they would probably reverse the causality: things break on TikTok and then become famous on Spotify. But that is a function of both algorithms and people pushing music, not necessarily only a function of the quality of the music.

13. Spotify, TikTok & YouTube Music

How do you think about the battle between Spotify and TikTok for music discovery?

They are both really important parts of that ecosystem.

TikTok wins on short-form video discovery, and Spotify wins on its algorithmic Discover Weekly playlist, which is much less visual and much more traditional, list-based recommendation.

I think that is right, but a lot of what becomes popular is also found on TikTok first.

How do you think about YouTube versus Spotify? That is the other big battle. There is long-form video and short-form video, and Spotify is trying to take on both at the same time.

In some sense, I hope Spotify wins because it would mean people were finding music more engaging.

YouTube is more engaging than Spotify, as measured by the amount of engagement and the ARPU for ads, for example.

Why do you think that is?

Video is engaging.

There is a ton of background listening to music, which on its face is not bad. I am not against background listening to music, but there is so much more that is possible with music than something you put on in the background to deliberately not pay attention to.

I will point this out to people: I interview a lot of engineers, and maybe 100% of the student engineers I have interviewed in the last 2 years listen to music while they code. That is great, but the whole point is not to pay attention to the music.

With Spotify doing video and making a big push, how do you think that plays out?

I hope people find it more engaging.

Let me answer the video question more tangentially. The vast majority of generative video companies have asked us for an API so they can have Suno music behind their AI videos, and the answer is always no.

Why?

There are a bunch of answers, but I think the reason is that we are trying to make music more valuable for people. Being the background music for your video is not making music more valuable.

This is not about cost. You can get Epidemic Sound music very cheaply, and that can be the background for your video.

What did you not do with Suno that you wish you had done?

A very obvious one for me is getting off Discord.

We released our first product in August of last year. In November, we put up a very thin web app, and I got this totally wrong.

I said we were going to be on Discord forever. I looked at Midjourney—they were printing money and were just a Discord bot, at the time—and I did not appreciate just how much a good UI would totally change the experience for people.

Discord is not the best UI for what we do. It is better for Midjourney than it is for music.

We released this web app, which did not even have the full functionality of the Discord bot, in November. It took 5 days for 90% of the traffic to move over to the web app.

There is no world in which you can say that I got that right.

14. What is Great UI for Mikey?

So you wish you had done the web app and the mobile app sooner?

Yes.

If you had done them sooner, would you not have cannibalized the Discord audience? Would there have been no reason for them to be there?

I would argue strategically that we now have 400,000 people still on Discord.

Nobody leaves Discord servers. Discord serves a completely different purpose now. It is a community, an amazing resource for us to talk to the community, and an amazing resource for us to get feedback from the community.

I think the product would have reached more people if we had built the web app sooner.

You mentioned the importance and significance of truly great UI. What is great UI for you?

I am always thinking that the UI should in no way resemble what it does today.

Good UI is about what you enabled that was completely impossible in the old pattern. What could I not do in a digital audio workstation?

It would not be easy for me to take a whole song and cover it in a different genre, but if I think, “That is actually something somebody might want to do,” I can treat that as a first-class workflow and build it.

That could be a really simple, beautiful, intuitive UI, but it would also be totally different.

15. How Important Are Prompt Guides for AI?

How do you think about prompt guides? It almost reminds me of knowledge management systems—the screen of death, bluntly. You see the box of death.

For someone who is used to prompting, you think, “I know to be specific and include very pointed directions.” For other people, that may be less important. How do you think about the importance of prompt guides?

Let me preface this by saying that OpenAI and ChatGPT are amazing. They did every AI company a huge disservice because everybody thinks that an empty text box is now the right interface.

It is right for ChatGPT, and it is incorrect for basically everything else.

I hate to be this guy, but I am going to reject the premise. I hope that in 6 or 12 months we are not using the word “prompt.” There are far more intuitive ways to interact with music, and we should not be guiding you; we should be listening to you, if that makes sense.

I had Bsky [?] and Gustav on the show, and they said we will measure the quality of a candidate by the quality of their prompts. Do you think that is not true?

I really hope that is not true. I might be wrong, but somehow this would be a huge missed opportunity if we cared about the inputs and not the outputs.

Should you not measure it by the quality of the music, or the quality of whatever it is they are doing? If it took a really complicated prompt, is that a good thing or a bad thing?

If it took a really complicated prompt and 1,600 iterations, that is a failure of my product, not a failure of the candidate.

I asked you what you did not do that you wish you had done. What did you do that you wish you had not done?

There is a part of me that wants to say, “Allow remote work.” That is a spicy one.

The thing is, every single CEO I know is dealing with it. I was walking in the park with the CEO of an enterprise software company this morning, and she said, “I know remote work is killing my company. I just cannot get people back.”

We never built the company remotely. We built it in person, and we make exceptions for people to be remote. It is generally worth it.

Is that not very hard, respectfully?

It is very hard.

Some people are worth it, but I think that as you scale, exceptions are a sign of judgment. Judgment—you can call it taste—is underrated in terms of how important it is compared with people’s skills.

As you scale, the judgment required to make these exceptions is very difficult. We are much bigger than we were a year ago, and from my seat I can see that this is already beginning to get difficult. Is that person an exception? Is that person an exception?

Can I ask you about the funding side? How much have you raised now?

A little over 125.

The last round came quite quickly, I think.

It came quite quickly.

How did you think about that? Was it important to cement your position as the market leader? Was it needed for GPU spend? What was the rationale?

I think about capital as a weapon.

You can deploy it, and there are step changes where more capital lets you do something differently. If you just added 10% more capital to your balance sheet, that would not necessarily change what you could do.

Do you think founders and VCs are aligned 99% of the time?

Yes, I do.

What do you know now that you wish you had known before starting this company?

There are lots of things I wish I had known.

I do not know how often you say this, but we reserve a good portion of our fund for capital later on in case you need it.

That is code for, “I am going to take my pro rata, and there is nothing you can do about it.”

In the good cases.

In the good cases.

In the bad cases, I am not going to give it to you.

Exactly. No fucking way.

That is built as very founder-friendly, but it is actually the opposite.

16. Scaling Revenue Beyond Other Generative AI Models

How have you scaled revenue in a way that other generative AI companies—image generation and other forms—have not?

I think caring about it is the answer.

There are no secrets here. We started charging from day 1, which may have been a happy accident.

I care much more about building a giant platform right now than I do about revenues, but I need to make sure that those things are somewhat commensurate.

You can come up with edge cases where I have to sacrifice all my revenues for the next opportunity—I have to do the Intel thing—but absent that, revenues growing is a sign that people are experiencing something they want to pay for.

That is a good thing, and we should not ignore it.

Has there ever been a time when a focus on revenue has led to the deprioritization of growth?

No, because we have always prioritized growth over revenue.

I think good people can keep both things in their heads. We want to grow, but we also do not want to sacrifice revenue.

If you hold my feet to the fire, maybe I can give you a quantitative number: I care about growth twice as much as I care about revenue. But where we are now in our journey, we are not really making trade-offs yet.

We have reached a tiny fraction of the addressable audience for the product as it exists today. We are building much more product, with much more valuable and fun experiences, that will reach an even bigger audience.

If I am making trade-offs now, I am doing something wrong.

How do you think about Suno being a social network? It is quite easy to see it going down the SoundCloud route, where you have profiles, feeds, followers, and likes, and it becomes much more socially interactive, with real social capital tied to status and profiles.

There is a bit of that.

Where SoundCloud goes wrong—through no fault of SoundCloud—is that it is still hard to make music. You are limited there.

It is also still the case that there are creators and consumers, and they do not really mix. That is unfortunate.

Music by itself is social. Is it social like Facebook? Is it social like the peer-to-peer elements or the long tail of Instagram? Is it social like the top end of Instagram, where you have people who look like they spend mid-6 figures on every post, make their livelihoods doing it, and have 100 million followers?

The answer is that it is all of those things and everything in between.

Instagram is a good example. It is crazy that the same platform lets me send something silly in a direct message and lets me consume content from a true professional who spends all day agonizing over a post.

Why cannot music be the same way? Music should not be special in that respect.

17. Should We Embrace or Fear Rapid Change?

I feel quite unnerved today. I feel unnerved at this time because I have never known the world to be in such a state of flux.

There are global conflicts and everything in between, and elections are changing the structure of countries so quickly. I also cannot predict the future of labor, content creation, content distribution, or information. Everything is up for grabs.

You can say what you want, but we have been creating music since before recorded history, and Suno can do that beautifully in seconds. It is amazing, but I feel uneasy that there is so much change happening so quickly.

What would you say to me? Should I be excited? Should I be nervous? How do you feel?

I feel very optimistic.

Let me say it like this. I assume you are talking specifically about music.

Some of it applies outside of music, but yes.

To whatever extent there are 2 sides to this—there is the old industry, and then there are the disruptors—both sides will end up agreeing on 1 thing. Everyone on both sides will say, “We know it is coming. We know it is inevitable,” meaning AI.

People in the existing industry say it because they do not want to sound like Luddites or hermits. They want to appear as though they understand it. People in the AI community say it because it is a security blanket: “Do not worry, I am not coming to disrupt you.”

But we know it is coming anyway.

In both cases, this is a really bad thing to say because on both sides it is essentially putting your hands up and saying, “It is coming. I do not even have to do anything about it.”

The future is ours to build. We can build a good future of music with AI, or we can build a bad future of music with AI. We can also sit back and let someone else do it.

How can the CEO of a public company, the CEO of UMG, say, “We know it is coming,” and not address the core of creation?

I think that person would say, “AI is an important part of the creation stack right now,” but would not try to address all the other people who want to create with these new tools.

I think it applies just as much to me as it does to the CEO of Universal Music Group. We can actively build a bigger, better future of music with AI, or we can wait and someone else—in another country, not bound by U.S. laws and without the same intentions—will build the worst future of music with AI.

I can think of lots of dystopian, worst-case futures of music with AI. One particularly bad one would be a group in another country that does not want to follow the laws making it possible for you to impersonate your favorite artist without permission and make endless copies—endless Ariana Grande songs, as you said—without giving consent to Ariana Grande.

That would not be good, and it is possible today.

Totally. You saw this with the likely Drake and The Weeknd Ghostwriter song.

It will be good, and it will just get better and easier.

That is not a good future.

There is another bad future, though. Music is meant to be social, but there is a local optimum where music gets less social and becomes far too hyper-personalized.

I would hate for the future of music to be that you open your phone, open an app, and hit play. It knows everything about you: what you did this morning, who you texted yesterday, and your mood. It has your Apple Watch and your heart rate, and it streams you endless music that only you are going to like, hitting that nerve in your brain like a drug.

That is extremely antisocial. Somebody might do that, and I think it would be a shame because it misses out on a lot more that is fun.

You remove all human connectivity.

There is an inherent tension between social stuff and personalization, and I do not want to make the hyper-personalized version the default.

You think about that because, as I said, I created a song for my girlfriend. Is that good? Do you want that?

That is fantastic. You made it for somebody.

That is much more social than what exists now, where you listen to music from artists, love the music, and see a piece of yourself in the artist, but do not actually interact with them.

You might say, “I wish you had spent 10,000 hours playing guitar and then you could have done that without the help of Suno.” But short of that, is it bad that you did it? I think it is actually great.

I do not want it to be only that. There is something about art where it is all about the artist. A lot of artists will not even care what other people think of the music. This is art for art’s sake, so to speak.

But if that stream is so pleasurable that you forget about everything else, I think that is not as big, good, or valuable as what could be.

18. The Balance Knowing When to Start and When to Stop?

You said the word “art.” Before we do a quick fire, I cannot believe I am quoting Ben Affleck, but he said at a panel—I am sure you may have seen this—that “art is knowing what to do; craft is knowing when to stop.”

How do you think about that? It reminds me of what Hemingway said: “Write drunk, edit sober,” or something like that.

I think where this all goes—and this is a very natural progression that is already happening—is that taste is increasingly the only thing that matters in art, and skill is going to matter a lot less.

You are going to be able to make a lot of stuff, and the people who are going to be recognized are the people who can pick from that vast quantity of stuff and, in the case of music, use their ears to say, “That was good, and that was bad.”

I think that is accelerating a trend that already exists.

Think about it: 30 years ago, if you wanted to be a rock star, you wanted to be an amazing guitarist who could shred. 15 years ago, you wanted to be a DJ, where you had to learn pieces of software, but maybe it was not quite as virtuosic as spending 10,000 hours playing an instrument.

Now people want to be influencers. They want to be famous for the sake of being famous.

In music, it is not really influencers in that way, but we will increasingly look at people who have good taste. You see this even with playlists. There was a group of people who made playlists—compilations of other people’s music—and it was, “I have good taste, and you are going to care to listen to my playlist.”

I think this is where it goes: “I have good taste. I make good Suno music.”

I cannot play the piano or guitar, but I am very good at picking through the Suno music and finding the way to make it sound the way I wanted it to sound. That makes me a creator.

19. Quick-Fire Round

I want to do a quick fire with you. I will say a short statement, and you give me your immediate thoughts.

Okay.

What do you believe that most people around you disbelieve?

That AI is inevitable.

What I said before is that it is coming.

I think everybody believes it is inevitable.

Why is it not inevitable?

We need to do it. If we just say it is inevitable, it is not going to come and happen. We need to go and do it.

Do you agree that we are going to go through a trough of disillusionment? Everyone is expecting AI now and next year.

I went to an event very recently with one of the biggest banks in the world and some of the biggest CEOs. The ROI is not here. We thought it was here, and it really felt like we were going down in terms of an AI winter.

I do not know. I live in a bubble in Cambridge, Massachusetts, and you live in a finance bubble in London.

Those people might experience disillusionment, but is the whole world going to be like that? I am not sure, because I think most people are still not expecting anything.

Should TikTok be banned?

My gut is no, but I am the wrong person to ask.

Why is your gut no?

I have a visceral negative reaction to banning things. I have a visceral negative reaction to banning things.

I just learned that Australia has banned social media for people under the age of 16. I do not think that is a good idea either.

These things are typically heavy-handed and do not necessarily accomplish what they are intended to accomplish.

Maybe that is a tactic to avoid the question. I am also not sure you want my opinion on that. You are a technology guy who does music.

What is the hardest element of being the CEO of Suno?

I do not know if this is special to us, but I think it is focus.

At any given time, there are 30 things we could do that would be very impactful for the company, and we have to choose 3 of them. We have to choose carefully.

The future of music is a greenfield right now. We could be attacking a lot of different things, and we have to be choosy about which ones we pursue.

What would you like to do but will not do for the benefit of focus?

Off the top of my head, everybody is asking for an API. Do we want to do that business? I do not know. That does not really build the future of music that I want to build.

There is always a tension between how many of the old power tools we want to build and figuring out what the new power tools for creating music are.

You said good judgment is more important than good skills, probably because we know how to assess skills and do not know how to assess judgment. Everyone is teaching for the test here.

How did having children change how you lead?

The first-order effect is that I have less time, so you have to be a little more ruthless with prioritization.

The flip-side answer is patience. When you see fast learners, you have to be patient with them.

Little kids are incredible learners. They pick things up so much faster than you and I do, and it is incredible to see. It is pretty special.

Tell me what concerns you most in the world today.

People are not good at realizing the first-order effects of the issues they deal with or the things they are doing.

People usually glom onto the second-order effects because sometimes they feel smarter, or it feels closer to home. But people are not good at judging effect sizes.

I will give you some examples. I teach a course at Sloan, at MIT’s business school, so you need to have an AI policy. It is a course, so are you going to allow students to use ChatGPT?

You get a lot of people saying, “GPT is the end of education.” My answer is, “No. The first-order effect is that GPT means every person in the world has a median-competent tutor or sidekick.” That is obviously amazing for education.

If you cannot recognize that fact, I am not sure you have very good judgment.

Let us talk about the second-order effects. Yes, all my homework just got made obsolete by one good prompt. My response to that is that, as the instructor, I now need to change what I teach people.

That is scary and a lot of work, but if I do not change what I teach people, I am not preparing them for the real world. Most companies are going to let you use GPT when you are doing your day job, so I should probably prepare people to do that.

The penultimate one: what is the coolest use case someone has used Suno for that you did not expect?

There is now a lot of usage with kids. Kids love music and resonate deeply with it. Half of my usage of the tools is with my 3-year-old, and I did not expect that.

There are lots of things we did not expect. We were at the top of the generative products that people put on their Ramp credit cards, which means that lots of people are using this for cold outreach, sales, marketing, and all kinds of things.

That is very cool and exciting. They are entirely new use cases, and it is very hard to say no to cold outreach when somebody writes you a personalized song. I imagine it is very effective.

I sent the Revolut team the song we did for likely Nik Storonsky, their CEO, and it was great.

I love that.

The final one: what question have I not asked you that I should have asked?

I think you asked it, but we got sidetracked. It is, “What should the future of music be?”

That is the question I think about night and day, more than any other question. “Should” is the important word because we need to build it.

The good outcomes here look like a lot more people doing a lot more music for a lot more hours of the day. All of the economics will flow downstream from that.

I think about AI as a tool for achieving many of those ends. It lets many more people participate in creation. It lets us recommend music to people more effectively. It will let you tailor a song for your girlfriend to her tastes, which is very powerful because it turns out she has to like the song for that effect to happen.

It is a world where a lot more people can make a living doing music. Right now, not enough people can make a living doing music.

I think about Instagram, or photography before and after Instagram. Very few people made a living doing photography, and now many more people can make a living doing photography because of Instagram.

Yes, it changes the tenor of photography. Yes, it means that every individual image is less valuable than it used to be. But in aggregate, this is a much, much better future, and that is what I would like to do for music.

Mikey, I have loved doing this. I love the product, and I have so enjoyed this discussion. Thank you for putting up with my wayward side questions. You have been fantastic.

This was so fun. Thank you.