Simile raises $100M from Index to build society-scale AI simulations — predicts 8 of 10 earnings call questions

lambda lightning round.

Let's bring in

I'm going to tell you about Vanta automate compliance and security. Vanta is the leading AI trust management platform. And we will bring in June Park, the co-founder and CEO of Similey. Welcome to the show June. How are you doing?

What up? Hi everyone. Great to be here.

Welcome to the show. Dude, you know you're you mean serious business because

behind you. I love the whiteboard. Uh is incredibly detailed. Please uh first time in the show. Introduce yourself. Introduce the company.

Yeah. So really excited to be here. Uh I'm Jun. I'm the CEO of Simile.

Simile is a company that is building foundational model of human behavior

that can simulate our society from ground up at the level of individuals and populations and predict its outcomes.

Interesting. We use that technology to create products that enable people to talk to millions of simileies we call agents. Yeah.

Uh that represent real people.

Yeah.

Uh I'm super excited about this. Uh when uh when we were working to get you on the show, I was thinking about it. There's been so many times in my career where I I was doing something like launching a product or launching a campaign or talking with a founder that was doing one of those things. And like often times you have a really strong sense of how something will do after you've built up some intuition. But I've always thought it was crazy that for some, you know, think of like a physical product company, like you have to spend a year and all this time and money uh and then just like launch it and be like, well, I hope I hope people like it. And it feels like with enough uh data and maybe a model like you're building it similly, you could get to the point where like you can get a much more accurate read on how society or a niche will respond to something without having to actually make the thing.

Yeah. What was the first thing you built for the company?

So for the company, we work with vendors uh that actually connect us with real human subjects. So we actually partner with people and get their data. And what we found was if we have rich data about individuals in our community, we can actually create really high fidelity simulation of these people. This was on the basis of the research that we've done at Stanford where we led agents and simulations based research. And what we are now finding is that this technology can actually create insights for our decision makers. So our core product basically is one that allows our Fortune 500 customers to actually interact with and talk to these agents that represent these real people that we collected

and basically provide insights. So that's the platform that we're creating.

Uh what's your thought on game engines as a piece of simulation? I've seen some economic research that maybe used Unreal Engine, but is there a role for uh simulated worlds or deterministic worlds that are then puppeteered by agents uh in any of this or is that just like a different fork of research entirely?

No, I think it's all connected. In fact, we actually started our research when we published the generative agents paper which introduced the idea of creating simulations and agents. We actually demoed this in a game town. One because we thought it was so visceral and so fun,

but also at the same time, if you want to create a simulation of our entire world and how our society might get shaped in the coming decades, then you really ought to be modeling not just people, but also the environment and the model as a whole.

So there's absolutely a place uh for these kind of game simulation like characteristics to this field. Basically the idea here is down the line can we actually create simulations of the entire market. entire nation. That's where we're headed.

So, yeah. What what what are some of the case studies that you've seen with uh large companies? Is it something as granular as should they charge $2 or $ 250 for a particular product or is it more broad? Like if we do a Super Bowl ad, will it will it convert or you know or just

Yeah, Super Super Bowl ads are such a good example where you're spending uh maybe a million uh 8 million on the buy

uh you know half a million on creative all this stuff and then

and then uh it'll either get people to hate you or love you.

But we should be able to predict these things better. And obviously historically there's focus groups you could try to get a reaction but we're still at this point where sometimes you just need to do something to see

indeed in fact uh as somebody who's running this company I actually simulated how today's conversation might go

really because a few of you are so well known online that we could actually get in there

to actually see how conversations like this one actually might pan out so I could be better prepared to talk about some of the topics that we wanted to cover. So core use cases today actually does cover things like so we work with fortune 500 companies that have earnings call many of them have come to us and said can you actually create simulations of let's say earnings call that's going to happen and there basically the idea here is can we simulate the each analyst who might actually ask questions during these earnings call the way to how CEO and CFO might be able to respond to these questions

that's a common ask so these are the kind of simulations that we create uh for that particular use case and what we find is we can actually predict eight out of 10 questions on average for these earnings score like the actual questions that will be asked and we can actually simulate how the conversation might flow from that point on. So you can test out different strategies.

Yeah.

But beyond that

one of our sort of customers right now is CVS.

CVS of course a large retailer.

Yeah.

And there CVS has been working with Simul to create simulations of hundreds of thousands of their customers. Mhm.

What they want to know is let's say CVS has new product, they have new store layout, they have new concepts they want to test. These are the kind of things that CVS today might go to a large human panels but instead they can now work with Simil to get insights that's much more granular and that's much faster to gain.

Now another example here Gallup sort of a longstanding polling company has been working with Simile to create digital panels of their human panel members. So if they want to ask questions to that panel but for whatever reason they are not reachable then you can actually come to similate to create simulations of that panel.

Mhm. What's the moat? How do you think about um like in the earnings call example? A lot of earnings health transcripts are already in pre-training corpuses and you could dump them in and sort of do a quick fine-tune or even just throw it in the context window. And if you're a public company CEO, you could probably just ask any of the frontier models. What do you think the analysts are going to ask? And is it that that strategy only yields two accurate questions and you're at a higher benchmark rate, or do you think that there's a different piece of your business that the big labs won't go after?

So fundamentally, the model that we're trying to create is different than what big labs might be interested in today. Okay? that we are less interested in creating intelligence that's superhuman, that's super rational,

but instead what we're interested in creating are models that actually model people's irrational half of their brain. These are people's values, preferences, taste. And the way we go about doing this is a true combination of modeling and product frontier. So, similar is unique in that we as a team actually is composed of both of those talents. So myself and my co-founders Michael Bernston and President Leang are all researchers who have been making contributions at the frontier of AI ranging from generative agents to generative AI and foundation models

and there the core mode here is what are the best kind of data that we can collect

and then using the data to create the most highfidelity agents that represent the people we have interviewed collected data from.

Mhm. And then we overlay that on top of amazing product.

So here one of the amazing part about the models that actually map people's values and preferences is the better model immediately means better insights for customers like CVS and Gallup.

So that there's an amazing alignment between the modeling frontier and product frontier which is not something that we see every day.

So it's a true combination of those two things coming together that makes sim special.

Yeah. what uh what what do you see progress scale with more? Are you compute constrained or is it uh much more about data constraints?

So we are trotting the frontier on both front. So down the line so today we model hundreds of thousands of people

who signed up to participate in these studies

and then our customers of course are they have appetite to down the line simulate millions. Our goal eventually is to ask ourselves what would it mean if we can create simulations of eight billion people the entire earth.

That's the future that we're headed.

Yeah.

So there certainly there's a question around what is the data?

Yeah.

But also how do we compute these kind of simulations efficiently?

Yeah.

Right.

Are you worried about getting too accurate that life just be kind of becomes boring because

is free will is free will real?

Do we have free will?

Are we living in a simulation? It's like you've read you you screw you went to the end of the book and you read the last page and you're just like ah like

yeah there's something fun about actually this discussion. Um so I don't know if you have watched the Matrix or I don't know how many of the audience actually have watched the Matrix.

I love the Matrix.

Amazing.

I have seen that one.

You have seen that one. Jord hasn't seen a lot of movies but he's seen the Matrix.

There's actually a quote that I actually like quite a bit that I think it's actually quite profound. It's something that the oracle says which basically is you're not necessarily here to make a decision. You've already made that decision. You're here to understand the decision and why you made that

and that sort of captures essence of simulation. Right? that the models that we're creating is isn't merely trying to predict the future on the we are good at that and that is one of the core capability but what to me is more important is understanding why we made the decision that we created and the creating these kind of bottomup simulation where we can actually go back and actually trace through the audit logs of how society might unfold

is actually an amazing way to gain that interpret interpretability layer of our reality

so that is to me the best way to shape the future Right? If you want to shape new policies, new product that would actually serve the people that is in our society, the best way is actually to understand those people

and then actually communicating with them at a scalable way

to basically create those policies. So that's how we view things. How much did you chat? Uh before before that, the chat wants to know, do you think uh do you think if you worked with uh Clvicular the live streamer, you would be able to predict whether or not he would get frame mogged if he went to different you know locations or or maybe universities like maybe you could have said don't go to ASU just don't just don't.

How much did you raise?

So we raised $100 million uh from index and other participants Uh

we're also really excited to invite angels uh such as Andre McCarthy.

They're very excited.

Okay, it's great big dogs. But ultimately what we're excited about is that there's a vision that I think is extremely compelling that this is a vision that this particular team have been working towards for the past five years and that we can create with the technology that we have today. Really high fidelity simulation

that can actually provide insights to create better society.

Well, thank you so much for taking the time to come chat with us. Congratulations on the massive round and all the progress and

great great to meet you. Very exciting

stuff. Keep tearing through the Fortune 500.

Don't Don't be afraid to send us some alpha. If you're predicting anything about us, let us know.

There's any I like this. I think you're starting a new meta. Somebody in the chat said it's kind of the Jensen Steve Jobs combo effect with the outfit.

I like it. Looks good.

It's It's great.

Well, have a great rest of your day. We'll talk to you soon.

The team.

Goodbye.

Cheers.

Let me tell you about Gemini 3 Pro. It's Google's most intelligent model yet.