The Exchanges

Q So talk to me. It was Char Dool that introduced us. Huge thanks to Char Dool for that. Um, but there's a new funding round that's come to be in the last month or so. Can you talk to me about the funding round, how it came to be, and how you think about it?

A For sure. So, uh, we raised our hundred million dollar round, uh, about five months ago. Um, and soon after, um, we were preempted, uh, fairly recently, um, by insiders. So Schrodo, led index, led our previous round, and including Schrodo and some of the other insiders were looking at the market. And Schrodo has sort of this comment that he every once in a while makes, Where he's seen some of the fastest growing market, and his track record does show that he truly has seen different markets. He has quite never seen this kind of traction, this kind of pull. When that is paired with the technological progress that has been made, and also the amount of compute that we can also leverage to even further accelerate our progress, that sort of prompted our insiders to go, can we actually put in more money now than later? Um, so that's how the initial round conversation came to be, and we were now planning on raising at that particular moment, but There are a couple of teams that I particularly respected in the Valley, um, that if we were to be raising, I wanted to talk to. And I found out that the team that I had in mind as my top of list actually was Neil Meadows team at Green Oaks. And turns out, uh, his team actually has been looking deeply into this market and all the players, how the market is going, and we're actually prepared to make the investment. And they're looking for sort …

AI assessment note: “we were preempted, uh, fairly recently, um, by insiders”

Q Before we move on to expansions that it could be used for, when did you know you had product market fit? You said you felt that pull. When were you like, ah, we got product market fit here?

A Many of the Fortune 500 board members and their C-suites reached out. In part, they do come to Stanford to see some of the demos that are happening in the lab, and they all saw the Smallville demo after it got released, and everyone thought, oh my god. If we can simulate a market like this, this is going to change the way we operate. So you could immediately sense the product market fit, and really this was the forcing function for us to then say, okay, this is actually quite interesting. We're actually going to show and validate that our simulation can not just be an interesting demo, but it's going to be accurate. So we spent about a year actually demonstrating that we can create models of people That are actually amazing and validated at predicting people's behaviors across surveys, behavior experiments, real environment, and we show that we can actually predict people's behaviors and attitudes 85% as accurately as people replicate their own. We put that work out at the end of 24, and that's really what started the field around synthetic panels simulations, and that's the market that we're seeing today.

AI assessment note: “they all saw the Smallville demo... you could immediately sense the product market fit”

Q ask you, when we think about the pursuit of some of the largest companies on earth, we mentioned, I'm not sure which customers were able to say versus not say, but you mentioned CVS. People always think it's like multi-year, incredibly long sales cycles. Was that something that you experienced, or was it a different experience for you getting and working with some of the biggest companies on the planet?

A What's been fascinating to me coming into the field of simulation, especially in this market, was Last year when I started the company, and when I, so I left Stanford in June of 25, so it's been exactly one year, I actually thought our field will actually, the market will take about a year or two before they warm up to the idea of simulation. So we'll basically find, build the right foundation for this company and for this market, and we'll go aggressive, maybe towards the end of the year, was what I had in mind. And that's not what we experienced. What we experienced was our customers were moving extremely fast, also in ways that, like, truly made me change my perspective on corporate America. Our leaders that we work with at, for instance, CVS, I've been working with this, Particular leader, Shree, who is their VP of Insights, extremely forward-looking, extremely ambitious, extremely hard-working, and amazing counterpart, two vision-like simile. But what I've also found was the pain they were feeling in their day-to-day work was so real. It was way more acute than I could have imagined. That when they realized that there is, or there could be an answer in this market, For addressing some of those pains, a very slow experimentation, budget, and so forth, they are ready to drop everything and try us out. So we actually saw some of the largest customers in the world move at ligh…

AI assessment note: “we saw some of the largest customers in the world move at lightning speed”

Q How do you solve that memory problem? Because everyone says, oh, we have a memory problem today. How do you solve the memory problem of agents to prevent that from happening?

A So back in the day, it was actually, the initial idea was fairly simple, which was that these language models are actually quite good at parsing natural language, so we put everything in Markdown text file. That was it. That sort of worked. Now, the issue there, however, is Because the language models have context window, and because, and even today, even if the context window is getting larger, the kind of experiences that these agents can have in this small game town is immense. And imagine now, if we were to bring this to real life in a world like the one we live in, the amount of memory that we accumulate is huge. So the problem becomes, how do you make sense of this large quantity of memory? So imagine you went to get omelet five types in a day. You want to make sense of that aside from, oh, I went to get omelet five times, you know, five times throughout the week or something like that. So we had this concept of reflection, which basically was every certain interval, it's like a shower thought. You have, you ask agent explicitly to get a bunch of their memory pieces and basically make sense of them. Why did you get omelet so often this week? Were you busy? Do you like omelet? Why are you studying for this test so hard? Like, you are in library every single day. Like, does this matter to you? And they will actually start formulating ideas that are more higher level than wh…

AI assessment note: “So we had this concept of reflection, which basically was every certain interval”

Q When you look at what can be done for some of the biggest brands, you mentioned like a CVS there, incredibly valuable for surveys, for customer feedback, for determining what customers really want moving forwards. I don't know how to say this. How do you, do you want to just be a next generation Qualtrics, and how do you prevent that being the end goal?

A Yeah. So the way we see it is, again, fundamentally, the core primitive of what we're, what we're trying to build is very straightforward. You tell us what population you're interested in, and we'll go model them. And really so far the layer of innovation has lived in the more tooling layer. How can we create better survey tool? How can we create a better interview tool? Simulation is fundamentally about something different, which is how can you create the most generalizable model of people so that we can represent people's viewpoints at scale? And that goes beyond simply running surveys or interviews. Down the line, I actually see simulation as a field moving into context where, hey, can we actually create simulations of many people interacting with each other so that you can understand all the downstream implications of your decision making? Or if imagine you have a new product you're about to launch, can you actually simulate the entire launch and how the audience might actually react, how the market might shift? And this also goes into the scientist part of me also get quite excited by the vision where simulation I do think can also be a cure for many of what we call quote unquote wicked problems. A good example here might be things like climate change requires collective action across many stakeholders who have different incentives. One of the reasons why such problems are…

AI assessment note: “Simulation is fundamentally about something different... goes beyond simply running surveys or interviews.”

Q Three months. Wow. Ok, that's very different to what people traditionally think. What matters more to them? Speed of output, in other words, being able to get results very quickly on their simulations, or accuracy of simulations?

A Yeah. It is both. There are so many questions that they are truly relying on their gut decision today. That if they can get some form of evidence to at least directionally guide them in the right path, then they're ready to try it. And then they very quickly realize that, oh, this is actually an amazing way to interact with a lot of data. This is amazing way to gain evidence that is actually quite accurate. And they actually, one of the ways we actually got some of our first customers was in the first call, They actually had a finding from, you know, large consulting companies, and they basically queried our system. Hey, if we were to rerun this, what would the system say? And we predicted the outcome of studies that took three to six months, but just within two minutes. That's very powerful.

AI assessment note: “Yeah. It is both.”

Q I love that. A lot of what people say is different to a lot of what people do. How do you think about the chasm of what people say, and what people do, and how that impacts your models?

A For sure. So say to give us real, and You know, if you look at the web data, it is fundamentally data of what people have said, not what they have done. And obviously large language models today are trained preliminary, um, mainly on this web data. For us, we actually do collect a lot of behavior data. We collect, uh, transaction data. We collect observational data. We also partner with our, uh, customers, uh, our vendors to collect some of this data. But my personal hot take here is a lot of observational behavior data. What they're amazing at is actually helping you create a correlation of the observation and what could happen in the future. Good for prediction task. But my take here after interacting with so many of our customers and also being in research, no one really cares about prediction. No one really cares about what's going to happen in the future unless you're trying to predict the stock market. What people actually care about is they want to shape the future. They want to know, imagine you're a Starbucks. Doesn't really help them to know that your Fraffino sales is going to tank in two quarters. They'll hear that and they'll be like, well, what do we do about them? That's terrible. What they want to know is how can we prevent it? What do we need to do now to change the future? And there, what you really need is causal mechanism. You need a model that can actually …

AI assessment note: “if you look at the web data, it is fundamentally data of what people have said”

Q How much data do you need to feel confident in an accurate prediction outcome to be displayed? Is it a hundred people? Is it a thousand people? Is it a million people?

A You want to have more people represented so that you can segment down to specifics of population. If you look at any social scientific literature, if you have a very narrow population of interest, You would usually get statistical significance in the study that you want to run by the time you have a thousand people. However, oftentimes the kind of ways that people query our system is they want to come in and say, hey, filter down to X, Y, Z population. Those filters are often created on the fly. For us to then be able to simulate people's responses across all those filters does mean we want to represent the entire population. So that's the journey that we're on.

AI assessment note: “You would usually get statistical significance... by the time you have a thousand people.”

Q Does it take a huge amount of compute to run these simulated environments at scale and well?

A Compute is an important piece of simulation. Of course, a lot of the work that we do is to make our simulation be more efficient. So a lot of our compute initially actually goes in to create the initial breakthroughs in technology. So it is actually exploring different ways to train, is exploring different kind of data set. Once we have a point of view, we can very quickly make it efficient. So some of the things that I've seen, uh, with Insimile, as we built this company over the year, is Right now, we have a model that's been in production. This model used to cost about a hundred times more to run than it does now. And some of it does happen because we actually found different ways to model, but with the same reward model, with the same philosophy, but just in a way that's much more efficient at inference time. So there are these kind of tricks that we can play and these kind of scientific advancement we can make to make things cheaper. A lot of the investment, however, does go to find that initial point of view.

AI assessment note: “This model used to cost about a hundred times more to run than it does now.”

Q When it comes to team building, you said about the craft of team building, and I think it's really interesting because it's an ongoing challenge building the best team. What have been your biggest lessons coming out of research in what it takes to build an all-star team similarly?

A So a couple of things. One is the team has to be balanced, that there are certain power that I can bring to the team, but there's also a lot of things that I don't know. I was a researcher. I was not an enterprise seller. I needed Lainey to be my co-founder to lead that part of the game. Um, so balancing the team and being able to see where your team is lacking, especially as we scale, there are new gaps that are emerging. Actually seeing that ahead of time and making sure that we fill those gaps, I do think is a core fundamentals of building a great team. At the same time, I also do think it's important that the team remains consistent in their values and in their rigor. This is more of a painter's analogy. Um, so when I was a painter, I was a figure painter, so I worked a lot with human subjects, portraits, figure studies. There's sort of this untold secret amongst figure artists, which is, it doesn't matter who you paint, your subject sort of looks like the painters themselves in some ways, or at least they share the similar vibe. I think building a team is actually a lot like that. In the, in the best team, in the team that you care deeply about, you really should see yourself in the team. And for me, a couple of things matters the most. One is, and this is the same standard I try to uphold for myself, but one is, are we the common denominator of success? People live throug…

AI assessment note: “One is the team has to be balanced, that there are certain power”

Q know the excess demand would be there. The paranoia drives the success. It's a really interesting one. Can I ask you, Brandon at McCaw was on the show recently, and he was like, honestly, researchers, they're in the tens of millions of dollars. It is, it is so expensive. Do you find that to be true, and how do you find this intense war for research talent in the Bay?

A Absolutely. So the research talent is very sought after today, and I have my closest colleagues and friends whose total com does range in tens of millions. Now, when they join, similarly, I, I'm fairly upfront with them. It is not possible. It doesn't matter how many hundreds of millions that you raised, um, meeting them at their base salary is tricky. However, the researchers fundamentally care about a couple of things. They care about a vision. If this idea truly come to fruition, like these are people who have literally seen OpenAI being the laughingstock at the, you know, in Silicon Valley to becoming a nearly trillion dollar business. And these are people who have seen Anthropik go to that same state within the past five years. So these are people who are fundamentally aware that deep, ambitious vision can actually come to fruition. So they care deeply about the vision. They also care deeply about the impact. What are the societal impact of the technology that they'll be working on, and is it actually interesting to them?

AI assessment note: “Absolutely. So the research talent is very sought after today, and I have my closest colleagues and friends whose total com does range in tens of millions.”

Q Do you worry about the retention problem in the Valley today? You see so many researchers move with such promiscuity, if you can use that word. Do you worry about the retention problem today?

A Consistently. And in fact, I actually do view the role of leadership to be and that of obviously hiring amazing people, but also providing a platform where individual members can express their superpower to their maximum degree. And I do think this actually part genuinely does matter. And retention can be challenging, but it can be done. And one of the core sort of a, I have a small sense of pride in the way my career has panned out over the past six years or so as a researcher, where PhD students often go from one project to the next, and their entire co-authorship will change, maybe except for your advisor. I've had sort of an interesting career where in the past six years, all my core team members never left. That we all move from one project to the next, to the next together. And now when I said, Hey, I want to do this thing and build a simile, I was able to somehow convince Michael and Percy who are, they were actually my doctoral advisors to actually come join me. And I think a part of it is, you know, we worked so closely together that there is genuine sense of trust. But at the same time, you know, I am somebody who fundamentally believes that one, it is my job. To communicate the degree of confidence and trust to the team that they feel like this will work out.

AI assessment note: “Consistently. And in fact, I actually do view the role of leadership”

Q is like, you know, if you look at, say, for Anthropoc and OpenAI and model routing, some tasks require, you know, frontier models which are much more expensive, much more token heavy, versus others which are much easier and can have a degraded or older model and a much cheaper model. Is that the same for simulations? Do different simulations cost different amounts in terms of compute token usage associated?

A They do. Um, usually when you have simulation that is trying to answer something that's much more complex, uh, or something that's, let's say you want to actually understand all the downstream implication of your decision, or you want to do market segmentation study across all of the US, much more expensive. What I also have seen, however, is in, it is in those simulations where we actually get higher ROI for our users. Because those decisions are some of the most costly decisions if they fail to make the right one. So this is actually interesting for simulation as a field. So you, we've all seen as a community, the inference costs going up and up and up, and we now have these thinking models that are thinking for like half an hour a day and actually start spending like token maxing and, you know, spending, you know, a lot of money on just running this process. I actually do think simulation can actually be the next frontier of that, where in my vision, I think there's a world in which in about two, three years, we're running a single simulation session, and that's going to take 10, twenty million dollars to run a single session, but it's going to be so valuable that people will pay a hundred million dollars for it. That's where I see it go.

AI assessment note: “They do. Um, usually when you have simulation that is trying to answer something”

Q Is data collection acquisition the hardest element of building simulation models for you? Like if we think about the kind of core pillars for traditional models, it might be compute algorithms and data. Is data the biggest challenge for you?

A Data is an important piece of Simile, for sure. Um, my fundamental thesis here is for AI companies of this generation, you need to have an interesting data strategy that's going to be defensible. And for us, really the data collection challenge comes from two angles. One is actually sourcing people. Sourcing people here is a little bit different than what other language model companies might consider to be their people or their population. We don't go after these expert programmers or expert scientists. We go after people like us, like everyday people living their everyday life. Um, that's, but what we care about is, are they representative? Do we actually have the same representation of people as we do in the world that we live in? And then actually asking the right questions to these people. What are the experiments? What are the questions that actually get at the fundamental core nature of who they are? Some of the questions we actually ask at the start of our data collection at times is actually saying something like, tell us the story of your life. Where did you grow up? What did you experience? What were some of the hardest problems that you had to tackle or decisions you had to make? Tell us a lot about these people. And that's what we try to do.

AI assessment note: “for us, really the data collection challenge comes from two angles.”

Q How do you balance the pursuit of the next dollar and serving customers who pay a lot of money, I'm sure, versus research prioritization and maybe focusing dollars there over building out a customer success team and an FD team How do you balance the profit maximization with the research purity?

A So Simile is interesting as a company. So Simile is a company that has a real product and engineering team, but at the same time, we are a research company. The co-founders, the four of the three co-founders are researchers. So we have as co-founders myself, Michael Bernstein, Percy Leung, Laney Allen. Myself and Michael and Percy were all researchers at Stanford. So I led research around agents, simulations. Michael was one of the co-authors of the ImageNab that really kickstarted the AI revolution, and he's been a leader in human-centered AI. Percy was the person who literally coined the term foundation model. And the vision for this particular area is the vision that we can actually create the next paradigm shift in AI and in the way we view technology and the impact of the technology in the form of simulation. The reason why we are able to, however, operate as a research lab, but also have an amazing product and engineering function and go to market function that's led by my counterpart Lainey, is the alignment between what the technology can do, the promise of technology, is so close to what our market actually requires. The better the model gets in representing people, the better simulation we can create, It immediately means better experience for our users, because they'll have much more grounded, much more accurate simulation. It is very difficult to maintain both a lab…

AI assessment note: “the alignment between what the technology can do... is so close to what our market actually requires”

Q Do you sit on top of core foundation models? How do you think about the relationship for those listening between an open AI anthropic frontier model provider and you?

A Yeah. So this is a great question. So the way we see it is if you look at large length model companies today, fundamentally the task they have at hand is to create super rational, intelligent machines that are good at coding, that are good at natural sciences and mathematics. Similarly doesn't really care about any of those. What we care about is if we have a person make a mistake in this context, we want our models to make the same kind of mistake. We want our models to be biased in the same way humans are. In a way, we want to be a representation of people's values, preferences, and taste. Sort of their subjective half of their brain. That's what we care about.

AI assessment note: “fundamentally the task they have at hand is to create super rational, intelligent machines”

Q In 10 years time, what is the craziest thing that you can see happening?

A A lot of things, but one thing I will actually say is I am someone who is fascinated by history of technology and analogies that we can draw from it. What I see today that's prominent in AI space is what I consider to be the CPU of intelligence unit. You have this one language model that's really large, that's very smart, that can do very complex reasoning tasks. That's like CPU. What I see coming and what I think simulation as a field can offer is the GPU of intelligence unit. As I mentioned before, Simile does not care about creating really smart, super intelligent machines. What we care about is creating models that are as smart as we are. I, I fed a lot of things. I want to make sure that the model that represents me fails the same way. But the beautiful part about people is individually, we have so much diversity, so much different takes in our world that makes individuals so interesting, but also when they come together as a large collective, the emerging phenomena that we're able to draw out is some of the most wonderful thing that we can see in our world. Creating a society, creating an amazing process that actually allows us to make all these achievements. Can we actually replicate that in simulation? I think it's going to be quite inspiring.

AI assessment note: “Can we actually replicate that in simulation? I think it's going to be quite inspiring.”

Q TAM in venture speak, you know, you obviously have your CVSs and your huge enterprises who would absolutely want to work with you. It can also be consumers, like regular consumers wanting to see what happens if, and running their own environments. Is this a play for everyone? Is this a play for the biggest companies in the world? How do you think about the TAM for something like Simile?

A So the start of my career really came from research, obviously, and the job of a researcher is to serve the humanity. Then we do our research, obviously for our own enjoyment as well. We love the process of finding new things in the world, but fundamentally it is a service. It is a belief that if we are able to make scientific breakthroughs, this is going to down the world, down the, down the line, serve everyone in our society. That is how I see simulation as a field as well. So right now we do serve enterprise customers for a couple of reasons. One, obviously, I'll be frank, there is the budget, that there is a clear product market fit that we see today, and that does excite us. And at the same time, it is an amazing way to validate the technology. It is very important to us that we get the feedback loop, To be as tight as possible so we know when our simulation's right, when our simulation's wrong, and we're improving it every single day. And obviously there's this side, you know, part here that's just as important, which is I have a colleague, ah, when I was at Stanford, ah, my office next, next to mine, ah, was Pat Hanrahan. He was one of the founders of Tableau. He's a graphics professor, also, ah, he won Turing Awards, a very well-known, ah, person in this, in this landscape. An advice he actually gave me and some of my colleagues was the best way to get feedback is to a…

AI assessment note: “So right now we do serve enterprise customers for a couple of reasons.”

Q If you were investing in my seat today, what part of the AI landscape would you say is under invested and most exciting? You can't say simulation.

A I fundamentally believe that, um, for AI companies in the future, you have to have interesting data strategy. Do you have access to data that no one else has access to? Do you know how to collect data that is very hard to collect? When you see those opportunities, I would invest. Uh, right now, aside from simulation, robotics is sort of an obvious place where this has become the case. Obviously robotics, there's a lot of money already going in. I wouldn't say it's under invested, but I also do think it is a quite interesting area. I also do think, uh, aside from the core sort of, um, robotics or AI space, the inference layer, but also chip layer, the hardware, I do actually think it's quite interesting. And it's a very hard area for people to crack into, but there are a couple of teams that have done, I think, an exceptional job in the recent months or years. I think they're quite interesting.

AI assessment note: “inference layer, but also chip layer, the hardware, I do actually think it's quite interesting”

Q What cannot be simulated today that you think will be possible in three years?

A So for me, it's actually a little bit less about what cannot be simulated, because I actually do think everything that we want to simulate, we can actually create the initial, uh, proof of concept. However, as we all know, one of the core challenges of AI is actually bridging the proof of concept with real value productionizable technology. So that's actually the chasm that I see. So interesting thing here is I see the world of simulation going into this world where we are creating that very complex multi-agent simulation, or we're running a very long study with many different steps of simulations along the way. But we actually started the field from multi-agent simulation when we created this small game town that was fundamentally that vision. And it sort of also makes sense because we did that because we, myself, Michael and Percy, we sometimes sit together and do this exercise called time machine game. If we were to write a time machine, go to 10 years into the future, what's going to be the craziest thing we're going to see, and can we do that now? And that was the motivation for running the Smallville experiment. So this can be done, but the question is, can we evaluate, ah, the efficacy of these simulations? Can we actually propose this as a scalable, productionizable system that people can actually rely on for making their decision? And that's the chasm. And that's the t…

AI assessment note: “it's actually a little bit less about what cannot be simulated, because I actually”

Q Can I ask you, thinking about that democracy's failing elections for a government, this would be an incredibly useful tool. How do you think about who you can and should work with versus who you shouldn't?

A Yeah. This is for us where the principles matter so much. The way I see it, simulation as a piece of technology is one of the twin pillars of technology. I'm a fan of science fiction. You read any advanced science fiction, there's always two pillars. One is some form of AGI that always shows up. The other is simulation. And like with any powerful technology, the misuse, the potential for misuse is quite real. And the way we see it, simulation at its best ought to be representation at scale. People have different viewpoints, different perspectives, different tastes. Many of their viewpoints are not considered in rooms where important decisions for them are made. We want to always say we listen to our people. We listen to our customers. We listen to our stakeholders. In practice, very difficult. This is a way for us to ensure that in every decision making, We actually listen to people at scale. That's the North Star.

AI assessment note: “simulation at its best ought to be representation at scale”