Grace Lee's Intelligence raises seed from Index to build a 5.6M-user design arena that trains AI on human creative preferences

Speaker 1: You get one Italian teenager to run this whole thing.

Speaker 2: They're like, if you were making $500,000,000, we would love to buy you for 1,200,000,000. Yep. But since you're making less than

Speaker 1: I don't I don't know if they're making any money. They must be making some money just as, you know, a website with some ads on it or something. Who knows? Anyway, let me tell you about Cisco. Critical infrastructure for the AI era unlocks seamless real time experiences and new value with Cisco. Our next guest is here already, I believe. We have Grace Lee from Intelligence, the co founder raising a seed round from Index Ventures. Brought you in a little bit. What's going on? How you doing, Grace? Good to see Great

Speaker 7: to meet you.

Speaker 1: Thank you so much for taking the time to come chat with us. Please introduce yourself and the company.

Speaker 7: Well, I'm Grace. I'm the CEO of Intelligence. We just launched yesterday. We've been the team behind Design Arena. Yes. Actually, the company started because we were making games. We're trying to make a game engine. We noticed that the models could make playable games, but the games were not fun at all. Like, nobody wanted to play them, and we thought, you know, building a harness probably doesn't seem like the right move. What if we could turn it both into a product for people to build games, but also a product for the models to figure out what it is that humans like to play, like what are playable games that are fun. And then we started that. We we just one of our friends linked it on Reddy. He didn't even tell us about it, and it got, like, 3,000 users overnight. We're like, well, that's more than the game engine has ever gotten. And since scaled that to, oh, man, 5,600,000 people across a 192 countries. So every country in the world is Central North Korea. We've been told that there are sanctions there, so so we can't go in there yet. And then on the Frontier Lab side, we pretty much work with all the major players to help improve their models and hard to verify domains.

Speaker 1: Amazing. Can you walk me through how Design Arena works, the various incentives that happen to actually get people to review designs for models and just the whole it feels like a product that would have sort of like a cold start problem. How do you solve that? How do you scale it? How does it function, like, in equilibrium?

Speaker 7: Oh, yeah. A really good question. So we we took a look at the optimal form factor for these cases where users are trying to do something Mhmm. But you don't really know what the optimal end end state is. So designers. Right? What do they show their clients? They actually they get the client prompt, and then they show the client multiple versions of maybe what they might want to see. That's exactly what the user interface looks like for our users. So they don't actually need an incentive because it's a design tool for them that does what they want because they show them multiple versions just like in the real world. Yeah. Like, how does a singer figure out if a song resonates? It's not that they cook in a room really I like to study their they just release multiple songs. Right? Yeah. They see what hits work. How does a a game developer figure out what games are are playable? They release multiple games and they see what works. So for the user, they actually don't need an incentive to be on the platform. They want to design something or make a game or make a website, and the form factor perfectly fits that. The the models, the intelligence doesn't exactly know what the user wants, but then they'll get multiple versions. And through feedback, you can get closer to the end state. And that just happens to be in the perfect shape that the intelligence layer needs to improve more to what it is that humans want in the first place, the preference signal, the the user behavior, etcetera.

Speaker 1: Little bit of, like, a, I don't know, philosophical question, but how do you think about optimizing design for the user and what the user thinks looks good versus what is good for the business? Because I've been like I've just been bombarded with products and websites that feel like they're horribly designed. I understand that they are economically valuable. John Gruber was on the show yesterday. He wrote a blog post recently about the Timu app being a mess. And if you go to Timu, it's like so many pop ups and spinning wheels. It is the epitome of terrible design

Speaker 8: Yes.

Speaker 1: In the aesthetic sense. But I'm sure every design decision that they've made has been economically justified and I don't think that they like never thought to make it clean, right? And so how are you thinking about those two tensions like actually instantiating themselves in the work that people do when they go to an AI model and ask for a design?

Speaker 7: Yes. Very good question. So, it is our job to figure out what better means for this person. Mhmm. If you are on the platform to make a beautiful portfolio, then you are going to give us feedback that hints us in that direction. Mhmm. Through a couple of interactions, we can figure out, okay, this person is actually just in it for the visual preference. Mhmm. Right? But if you are on the platform to build a highly performing website, like a team website, for example, you're probably optimizing for dollars converted at the last screen.

Speaker 1: Yeah.

Speaker 7: And so the end signal that you're tracking is different. The way that we like to think about it is we're almost doing the work of a PM. Right? What a PM does is they sit in the product, and they took a look at the analytics of what people are doing, and they have an end goal in mind. The PM's job is to figure out what is the right KPI to be optimizing for, and then to bring that to the table of engineers and and to the resource allocation and to try to get it in that direction.

Speaker 1: Mhmm.

Speaker 7: So it is our job to figure out for this particular user, what does better mean for them? And then measure those bright signals to then label correctly and then provide that as feedback to improve the models.

Speaker 1: How do you Jordy, please. You got a follow-up.

Speaker 2: I have a I have

Speaker 1: Okay. Yeah. Just following up on on Design Arena. How are you balancing the various downstream business opportunities that come from having a product platform like Design Arena? Because I could imagine you can sort of become a model router in many ways. You can do forward deployed work and be an expert and partner to companies that are picking tools. Like, there's services as a software. There's going down the financial side. Like, there's so many opportunities. It must be hard to like, you're sort of stuck in the idea maze longer than usual even after you have traction and fundraising and revenue. There's more opportunity. What have you looked at? What's not interesting? What has stuck out as the obvious path?

Speaker 7: Oh, my. Okay. That is the best question, by the way, to be asking in this space. Like, there are so many downstream opportunities that come from figuring out what it is that people want. What we found to be the most interesting, actually, is an intelligence marketplace.

Speaker 1: Okay.

Speaker 7: Right? There are some, this is a little bit kooky, but I do believe that there will one day be as many suppliers of intelligence as there are currently suppliers of information. This feels like the early days of the internet, Like, where anybody can make their own intelligence. I don't care if that's a foundation lab with a main API or something that's post trained or something that's totally off the shelf with a skill, but there's like a million different suppliers of information. And if you can be the person that best matches the supplier of information with somebody supplier of intelligence, I should say, with somebody who's asking for intelligence the same way that PageRank for Google search was the critical bottleneck to get the Internet into everybody's hands. That is the position that we actually wanna be in. So whatever the work it is that we do, we want it to fit into this flywheel that we have, and we want to make sure that we're not just good at value creation, like not just good at improving the models, but also good at value cap value capture. Mhmm. Once the models get better at design, it's not like our work becomes less useful. Mhmm. It actually means our product can reach more people because they're better at design. So we care about doing things in, like, the opportunities that are coming our way, which we're lucky to have a ton of inbound from from yesterday, that it feeds into this overall flywheel. But if the models get better, we also get better.

Speaker 2: Yeah. Jordan, please. Almost $60,000,000 of revenue. I think I think a lot of people were quite surprised, not because the company isn't like significant and at the center of this, you know, explosion of creative intelligence, just because it's you guys had kind of flown under the radar. So first, I wanna hit the gong for 60,000,000 ish of revenue. There

Speaker 9: you go.

Speaker 2: It's a great hit. Thank you. A very in a very like I like the ratio of the fundraise to the revenue, you know? Yeah. Normally when somebody comes on with this much revenue, they're like, you know, raising a billion. I'm sure that's up next for you. But I wanted to ask like what what do you think creative super intelligence looks like? Because with, like, images, I think I think images are gonna be solved. Right? At least, like, photo real images are pretty much here. You can imagine with a couple more turns, like, at least making images with AI will be solved. And Yeah. It's getting a lot easier to prompt, and then harnesses will get better at at editing and modifying and getting better outputs and all these things. But, like, photo real images will be solved. What does, like what does what does creative intelligence look like in one year or two years down the road?

Speaker 7: Creative superintelligence to us is whatever you can think will exist. In fact, things that you can't even have the scope to think of can exist. So, one one thing that we like the the distance between an idea in your head and it being in the real world right now is, like, very far. Let's say you have an idea for a new pen or what a new pen might look like. There's a very long distance that you have to travel. You have to know how to design the pen. You have to know how to process CAD. You have to find a manufacturer. Have to build like, there's a very long there's a very long and because of that, we have to localize on a couple of pen form factors that everybody uses. Like, it just it becomes standardized. Mhmm. But what if creative superintelligence makes it such that you can sustain infinite varieties, kind of like how you can have infinite versions of the same information being represented in in different formats? That's the way that we view creative superintelligence, which is any idea that you have, the instant that you have it, it is visible in the real world. And there's no latency and friction between your ideas and and them being born in the room.

Speaker 1: That sounds different than the models coming up with the ideas, which is interesting. I I I don't disagree. I think you're I think you're right. We were debating this yesterday about will a model be

Speaker 2: Yeah. Yeah.

Speaker 9: Yeah. Think I think have question.

Speaker 1: Before you even think of it? What is the timeline for that? That's

Speaker 2: Yeah. Yeah. The next thing? Yeah. We we were talking about this because it feels like, you know, mathematicians have been having somewhat of an existential crisis for for a while now, but especially over the weekend. Yes. And and I was just thinking, I was joking to John yesterday just saying like, you know, they're coming to us. Right? Us

Speaker 10: next time.

Speaker 2: Like, we're like like, I'm I'm I'm, you know, I consider myself a creative person.

Speaker 1: Yeah.

Speaker 2: And right now, when I look at AI generated outputs, the ones that really resonate are ones where there was like a fundamental creative idea.

Speaker 7: Yep.

Speaker 2: That then the model did a good job of instantiating it. But it's not like the original idea was that complicated. The example we always use is like Harry Potter Balenciaga. Right? Like, it's not a complicated idea. But to me to me, what will be interesting, and I'm surprised no one has like really built some type of like effectively like loop around this Mhmm. Which is like Yeah. Combining two random ideas and then creating We're the visual sort of output. And just doing that over and over and over because it will end up mimicking like the full pipeline of human creativity which is like original idea which is combining two things, Harry Potter, Balenciaga. Combining these two things and then just creating the output. And then once you get that, then you're just running on this loop where already it feels like there's no there's no real new ideas in the world. Like for Yeah. Like a hundred years, we've just been as humans, like recombining different things that already exist. And it's sort of sad that we can't come up with, very many new ideas. But then eventually, the machines will have just created every possible variation of every idea that there could ever be. And a human will have an idea and and and then they'll be like, oh, well, actually, the machine, you figured this out like two years ago and here it I is on a

Speaker 7: mean, I think something that's pretty incredible in in, you know, is people don't know what they want until they see it.

Speaker 1: Mhmm.

Speaker 7: Right? And oftentimes, it's not even the new idea that is like the the it's showing it to a dozen people and then two dozen people and then a 100 and a thousand and seeing what sticks. Again, this is kind of the creative process. Right? Like, you you will draft up a 100 version versions of of the same script and then see what are the gem moments that actually resonate with people. And the resonating with people, like, that's the verifier part. Like, that's where humanity comes in. Right? Like, if you're trying to give something to people, you need a way to measure whether or not it actually it actually resonates with them.

Speaker 2: That said, companies similarly who just raised Yeah. A big round might make, humans unnecessary to even judge what what will resonate with I

Speaker 1: think we're safe. Every AI lab I talk about is working on this thing called super intelligence. I haven't met a single AI researcher working on room temp intelligence. And so like they're not even trying to displace me. It's ridiculous.

Speaker 2: Actually there's probably a Neo lab called normal intelligence.

Speaker 1: Probably. Probably. Last question. Gaming arena. We got design arena. You have gaming DNA. What about models building games? The games go up against each other. Which one did you play longer? Which one did you have more fun? This feels like the next thing. Is this you? Is this another company?

Speaker 7: What's This is that.

Speaker 1: That's you.

Speaker 7: This is that. So we've actually been working on this for a while. Screw it. Thought Here

Speaker 1: we go.

Speaker 7: Game dev is incredibly hard to verify and that's Yeah. Why it's so much fun. Right? Because you don't just need a really good storyline. You need a way to verify if something is fun. Yeah. Possible to do, you know, without people on the other side. Exactly.

Speaker 1: Yeah.

Speaker 7: And so for a while now, we've been we've been helping companies try to get better at at this domain, and it doesn't just involve, like, the strategy and then being able to save game state. You've also got to have beautiful graphics. You have good multi processing. It's a very hard frontier.

Speaker 1: Yeah. Yeah. That'll be a lot of fun. Yeah. Yeah. Very very interesting to think about like, oh, I'm going to go and review my models on gaming arena. See you in four hundred hours because I need to decide which Yeah. Game is better. This hundred hour game or that hundred hour game?

Speaker 7: Yes. Yes.

Speaker 1: That's kind of what's gonna happen. Yeah.

Speaker 7: Yeah. You literally just need to put it like the way to think about it is you put it on the equivalent of the internet and you need to see if it gets picked up by people.

Speaker 1: Yeah. Yeah. Like the algorithms, like the Steam store, the Instagram Exactly. These are the final bosses of the arenas perhaps in many Exactly. Very interesting to think about.