# The Startup Building World Models From a Billion Video Game Clips

Markdown copy of https://realalpha.app/v/zsgnonEeAiQ/ on Real Alpha, the best minutes from long interviews with the people building AI. Cite and link the page, not this copy.

Pim de Witte argues that diverse gameplay data can pretrain agents for games and robotics, with targeted real-world fine-tuning.

- Show: Newcomer Pod
- Hosts: Eric Newcomer
- Guest: [Pim de Witte](https://realalpha.app/builders/pim-de-witte/) is CEO of General Intuition.
- Published: Aug 3, 2026
- Length: 59 min

## The best minutes

- [17:04](https://realalpha.app/v/zsgnonEeAiQ/?t=1024) [Game data can pretrain robots, but real-world tuning still matters](https://realalpha.app/s/game-data-can-pretrain-robots-but-real-world-tuning-still/) (9 min)
- [32:42](https://realalpha.app/v/zsgnonEeAiQ/?t=1962) [Specialized robots may beat humanoids in homes](https://realalpha.app/s/specialized-robots-may-beat-humanoids-in-homes/) (4 min)
- [42:26](https://realalpha.app/v/zsgnonEeAiQ/?t=2546) [Build scientific skills alongside AI literacy](https://realalpha.app/s/build-scientific-skills-alongside-ai-literacy/) (5 min)

## Also worth watching

- [36:10](https://realalpha.app/v/zsgnonEeAiQ/?t=2170) [Medal matches game skills to paid AI-training work](https://realalpha.app/s/medal-matches-game-skills-to-paid-ai-training-work/) (6 min)
- [12:28](https://realalpha.app/v/zsgnonEeAiQ/?t=748) [A mobile game can have sound mechanics and still stall](https://realalpha.app/s/a-mobile-game-can-have-sound-mechanics-and-still-stall/) (3 min)
- [55:34](https://realalpha.app/v/zsgnonEeAiQ/?t=3334) [Claude found an S3 key during reverse engineering](https://realalpha.app/s/claude-found-an-s3-key-during-reverse-engineering/) (3 min)
- [27:28](https://realalpha.app/v/zsgnonEeAiQ/?t=1648) [A top game using General Intuition models could arrive next year](https://realalpha.app/s/a-top-game-using-general-intuition-models-could-arrive-next/) (3 min)
- [5:55](https://realalpha.app/v/zsgnonEeAiQ/?t=355) [Factorio and RuneScape are strong hiring signals](https://realalpha.app/s/factorio-and-runescape-are-strong-hiring-signals/) (3 min)

Skip the opening RuneScape history and the long debate on AI demand, Chinese models, and distillation before the coding-agent security anecdote.

## Full analysis: Pim de Witte says roughly a billion uploads could aid robot training

- Medal gets roughly a billion video uploads yearly, and some recordings carry captured controller actions.
- De Witte says game pretraining can cut robot tuning to minutes, but real-world data remains necessary.
- The test is whether action data transfers safely across tasks, not whether the upload count sounds large.

Its 6 sections, what it means for you and what to watch are on [the page](https://realalpha.app/v/zsgnonEeAiQ/#analysis), free with an account:

- [15:12](https://realalpha.app/v/zsgnonEeAiQ/?t=912) A billion uploads are not a billion action examples
- [18:40](https://realalpha.app/v/zsgnonEeAiQ/?t=1120) Game pretraining may shorten, not erase, robot tuning
- [19:14](https://realalpha.app/v/zsgnonEeAiQ/?t=1154) Clip diversity does not prove real-world coverage
- [24:28](https://realalpha.app/v/zsgnonEeAiQ/?t=1468) Safety still depends on controls outside the model
- [35:39](https://realalpha.app/v/zsgnonEeAiQ/?t=2139) De Witte expects practical robots to beat home humanoids
- [54:13](https://realalpha.app/v/zsgnonEeAiQ/?t=3253) Data permissions and tool access need explicit controls

## Summary

Pim de Witte built the world's largest RuneScape private server as a teenager, then founded Medal, a gaming clip platform that processes roughly one billion gameplay clips a year. He is CEO of General Intuition, which uses game recordings and player actions to train models for robotics, world models and autonomous agents.

De Witte says gameplay recordings pair video with player actions, giving General Intuition data to pretrain agents across many digital environments before adapting them to robot tasks with minutes of fine-tuning. He argues that game data can support spatial and temporal reasoning but cannot replace real-world data, and says action models need safety checks. He expects AI-driven game characters and task-specific robots to benefit sooner than household humanoids, while saying gamers' distrust reflects botting, artist displacement and job concerns. He sees demand for AI growing but flags infrastructure, energy, competition, data access and model security as open challenges.

### Minute by minute

- [1:45](https://realalpha.app/v/zsgnonEeAiQ/?t=105) **RuneScape became his first business.** De Witte says he still uses Rocket League to gauge his mental sharpness, though he plays far less than before starting General Intuition. His RuneScape private-server business generated about a million and a half a year in revenue by age 18 and held about 60 percent of that market. He says he built the server by modifying the Java-based game, then shut it down after the company behind RuneScape became stricter on private servers.
- [5:55](https://realalpha.app/v/zsgnonEeAiQ/?t=355) **Games reward different kinds of skill.** De Witte says games with changing situations and self-directed goals can build long-term planning and discipline; RuneScape also taught him English. He names Factorio and RuneScape as strong hiring signals, with Factorio the stronger of the two, followed by StarCraft. He puts Rocket League lower because he sees it as relying more on motor skills than transferable strategy.
- [9:07](https://realalpha.app/v/zsgnonEeAiQ/?t=547) **Humanitarian work shaped his startup path.** De Witte joined Doctors Without Borders during Ebola, building medical-record systems and hardware, then satellite-based heat maps for disaster response. He helped create MapSwipe, which he says helped millions of people get mapped, but says its $50,000 budget had to cover four people's salaries for six months. Frustrated by the limits of nonprofit funding, he started a mobile game studio and later pivoted to Medal after its brawler lacked session variation and player liquidity.
- [15:15](https://realalpha.app/v/zsgnonEeAiQ/?t=915) **Gameplay actions can pretrain agents.** De Witte says Medal records gameplay clips alongside controller movements and button presses, allowing General Intuition to train models to predict actions from frames. He compares this to language models predicting the next text token, and says diverse game data supports spatial and temporal reasoning across games, computer use and robotics. He says models pre-trained mostly on game data have transferred to novel robot tasks with minutes of fine-tuning, while stressing that both game and real-world data are needed.
- [22:13](https://realalpha.app/v/zsgnonEeAiQ/?t=1333) **Negative examples support safety checks.** De Witte says gameplay that includes harmful outcomes can help models learn cause and effect and evaluate what not to do; he says AI labs wanted the data for alignment. He describes pairing action models with deterministic or sensor-based systems such as LiDAR to invalidate unsafe actions, alongside evaluations in harmful scenarios. He says General Intuition may open more negative-event data to other labs.
- [25:30](https://realalpha.app/v/zsgnonEeAiQ/?t=1530) **AI must preserve game integrity.** De Witte says General Intuition's agent plays many games at a high level and beats him at most of them, though not Rocket League. The company limits access to avoid bots undermining game mechanics, and he argues that AI companies should work with developers and disclose when players face bots. He expects AI-driven non-player characters to be an early showcase, with promptable behavior making games more interactive.
- [31:07](https://realalpha.app/v/zsgnonEeAiQ/?t=1867) **Purpose-built robots may beat humanoids.** De Witte says some robotics demos use sped-up footage or controlled stage settings, and points to Generalist as a company that shows its models' generality. He argues that humanoids are bulky and harder to train, while simpler robots built for specific tasks may be more practical. He favors wheeled robots with cameras and two arms, and expects humanoids sooner in factories than in homes.
- [36:10](https://realalpha.app/v/zsgnonEeAiQ/?t=2170) **Gamers want AI to solve real problems.** De Witte says gamers' skepticism is warranted because AI has brought botting and efforts to replace artists, while game studios face shutdowns. He says General Intuition wants to help with uses such as anti-cheat rather than replace art workflows. The company launched Nerve to match people with work based on action sequences they excel at; for example, it could use Factorio players to label factory-specific data.
- [42:37](https://realalpha.app/v/zsgnonEeAiQ/?t=2557) **Science and coding prepare workers.** For young people, De Witte recommends math and physics for rigorous problem-solving, plus coding and AI as basic literacy. He says biology has major potential, but scientific progress is constrained by missing measurements and data, making measurement tools an attractive career area. He cites Stanford's Eterna, where players used puzzles to help predict RNA structures, as an example of games contributing to science.
- [47:21](https://realalpha.app/v/zsgnonEeAiQ/?t=2841) **AI demand hinges on infrastructure.** De Witte argues that better models generate new questions and support more agents, so he sees no sign that demand for intelligence is capped. He says the economic outcome depends on infrastructure, energy and margins, and warns that lower-cost providers could erode Western model margins. He says open-source models can help General Intuition because it has a unique dataset, while reliance on Chinese models raises supply-chain concerns.
- [52:59](https://realalpha.app/v/zsgnonEeAiQ/?t=3179) **Data access and security need safeguards.** De Witte says he understands the emotional objection to using users' clips to train models and wants Medal users to share in the benefits through job or game-development opportunities. He says distillation is difficult to prevent, so labs need to be careful about what they train and how they evaluate models. He recounts an intern's legitimate reverse-engineering task in which Claude found an exposed S3 access key and retrieved SDK documentation, illustrating how AI can surface security risks users may not recognize.

## Transcript

Newcomer Pod, hosted by Eric Newcomer, with Pim de Witte.

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

[0:00](https://realalpha.app/v/zsgnonEeAiQ/?t=0) On RuneScape, I built the largest private server in the world, so that's the only one where measurably, I think, did pretty well. Is there money in that? Yeah. The company generated about a million and a half a year in revenue by the time I was 18 years old. You're saying there are more car crashes in the video games than in the real world? Yeah, we've counted. The amount of clips uploaded with car crash labels surpasses the amount of physical car crashes that happen in the United States every day. Are there games you're prioritizing to show off your capabilities? I think we'll leave a bit of a surprise.

[0:26](https://realalpha.app/v/zsgnonEeAiQ/?t=26) At 18, Pim DeWitt was running the biggest RuneScape private server in the world. Today, he runs Metal, the app gamers use to clip their best moments. They upload about a billion clips a year, and every clip records the exact buttons the player pressed. That's what his new company, General Intuition, trains AI on. The result is a model that reads a screen and works out what to do next. And that skill transfers straight to robots. Pim says his AI already beats him at almost every video game. almost, not Rocket League. This is my conversation with Pim DeWitt, CEO of General Intuition. Pim DeWitt, CEO of General Intuition. Thank you so much for joining the Newcomer Podcast.

[1:15](https://realalpha.app/v/zsgnonEeAiQ/?t=75) Thank you for having me. We have so much to talk about. We'll get into some of the big questions of artificial intelligence progress. You're building in what might be, you know, one of the buzziest categories in Silicon Valley right now, world models. But I thought I'd start off in more fun, lighthearted territory. You were at the top of your field in RuneScape and Rocket League, I believe. And they sort of helped you get into entrepreneurship, right? So yeah, tell us about your RuneScape days and are they all behind you yeah um runescape uh no i still play both games occasionally actually so i got so good at rocket league that i can kind of tell how my mental state is by how i play and whether i should be making important decisions that day so if i if i'm on it and sharp i can see that i'm playing really well i can essentially measure my my level of decision making by uh like if i play like a few games in the morning i kind of know whether or not it's like a good day to make big decisions and do you do that you're like i have a board meeting today let's play some rocket league and see if i'm on on admittedly my best admittedly recently no um uh but uh you know i think uh you know in the past year uh you know i i probably get 10 percent of the games in

[2:33](https://realalpha.app/v/zsgnonEeAiQ/?t=153) before i started the company uh but i don't know runescape i built the largest private server so in rocket league i was um i got up to i got up to gc1 so definitely not the top of the field but it's it's a it's a good place to be you know what top percent what percentage i don't know if i guess so it's double so i'm guessing top five percent or something like that i don't know what the today's numbers are um on runescape i built the largest private server in the world so that's the only one where measurably i think did pretty well is there money in that yeah yeah company generated about a million a half a year in revenue by the time i was 18 years old um and we owned about 60 percent of the private server market on runescape yeah is that still running no uh shut it down when the company behind runescape got sold i believe in 2015 it got sold to a chinese mining company at some point and like physical mining or crypto yeah there was the no physical mining yeah there was some there was some program at the time where the government was uh encouraging non-technical companies to go buy overseas technical companies And I believe they somehow ended up buying the creators of RuneScape.

[3:39](https://realalpha.app/v/zsgnonEeAiQ/?t=219) And then, yeah, as a result, we had to shut our servers down. Because they made everything private? Yeah, I think they were more strict on private servers than the previous ownership. Did you, you know, were you one of these people, like, harvesting in-game items and selling them to you? So you could, so in the private server, so I built a private server, which basically means... It means that you host and distribute your own version of the game, both the client and the server, which is fairly easy to do on RuneScape because it was a Java-based application, so you could essentially just decompile and de-obfuscate the code, and you could modify it, and then you could see exactly the interactions going to the server, and then you could essentially construct a server that acts in the way that the client would expect, and there was a whole community of people that was doing this.

[4:26](https://realalpha.app/v/zsgnonEeAiQ/?t=266) it wasn't just me it was like probably a thousand people that were all excited about this stuff um i just happened to create one that became very large um and yeah item spawning was one of the things that people made money on um but are you breaking the game like are there people who object to this for sure yeah this is on the edge of allowable uh behavior the private server got extremely popular after runescape removed the wilderness which was um sort of their primary draw for a lot of the like player versus player oriented people that played the game um and then yeah i think you also asked a lot of people come with runescape origins and i think the reason for that is is because runescape uh was free and world of warcraft was paid so all the rich kids played world of warcraft and all the poor kids or people that didn't want to pay the money uh played runescape and they they had to in order to get to the top you had to get the hustle you couldn't pay for it and so I think it basically self-selected into uh the people that actually wanted to because like RuneScape notoriously did not have pay-to-win mechanics right so you couldn't buy your way into stuff there wasn't like you could pay for membership but everything else you had to grind

[5:34](https://realalpha.app/v/zsgnonEeAiQ/?t=334) and so it was both free so it self-selected into the people that actually just wanted to work hard and then it actually was a very very great simulation of economies like I learned English there like I have my some of my colleagues that I played with still remember when I didn't speak can work at word of english so it teaches you in a very sort of general way um how systems work so your language started out in the most internet forum way possible yeah yeah i'm sure there are lots of parents you know they're yeah like oh my son plays too many video games oh there's a path to being an entrepreneur like what is is there i mean you know you took a very particular set of steps yeah to get to where you are like is there anything generalizable there yeah yeah i thought about this question a lot um yes i think as long as the games that you're playing don't have the same repetitive loop every game but there is a lot of variation so like in rinscape every time you play there was new economies new people um uh you kind of created the game yourself right because you got into the game then you decided what you wanted to do that day whether you wanted to go mining or questing or whether you want to go training um or whether you wanted to go

[6:39](https://realalpha.app/v/zsgnonEeAiQ/?t=399) develop like a long-term skill which would help you like agility which would help you advance on like moving down the map faster later but there was a lot of sort of long-term planning and discipline involved because like nobody wanted to click the buttons for agility but like they would get you much further later down the road so it actually taught in a very general way some of these more long-term discipline things delayed gratification delayed gratification yeah and so So I think all of these, so I think the answer is it really depends on which game, like if my, you know, if, and also I think the community that they do it in, like within games, there's always different meta games, right?

[7:17](https://realalpha.app/v/zsgnonEeAiQ/?t=437) So like, for instance, even in Rocket League, you can build maps. And if you're building maps, it's probably really good for you. All right. So yeah, top three games that are positive hiring signals. So I would put RuneScape and Factorio number one.

[7:29](https://realalpha.app/v/zsgnonEeAiQ/?t=449) **Pim de Witte:** i found factorio to be even higher signal than runescape and then probably starcraft uh is probably number three and then uh unfortunately for me rocket league is at the lowest yeah i was

[7:43](https://realalpha.app/v/zsgnonEeAiQ/?t=463) gonna ask the bottom why is anything where it's just your main skills like the controller input it's it's no good or you know maybe i have purposely avoided finding out the answer to this question at risk of uh uh no i think it's just because it's mostly motor skills and there's not a lot of, there's not a ton of strategy that transfers except for a few things i imagine like candy crush has got to be pretty bad anything where you're rewarded for just sort of checking out yeah my guess is it's something downstream from action space the amount of actions per minute you're expected to input into the system the the variance in the frames in the game the number of states so like ringscape has like an infinite amount of states because you could you could set goals those have sub goals and then there's like a condom like they need to go assemble like the amount of coins to get there there's like um there's over um like 20 000 items like so the amount of ways that the system can interact in runescape is so high whereas in rocket league every single map is kind of the same and and so it's it's just a very i play rocket league to relax if that makes sense um but i play like runescape if i'll go like decaying into the wilderness i want to be in it right uh and so

[8:54](https://realalpha.app/v/zsgnonEeAiQ/?t=534) think people just play for different reasons so you know for kids i think the most important thing is which games are you playing um and if your kid's playing rocket league go get the adoption papers yeah i love it and so metal yeah explain your sort of first foray and is that your first company or obviously the servers is that your first startup are there other startups too um so i started uh so my private server was my first company right which did yeah did very well um i I joined Doctors Out Borders after that. What is it, sir? MSF, the medical charity. I joined during Ebola. The reason was I love doing hackathons.

[9:32](https://realalpha.app/v/zsgnonEeAiQ/?t=572) So I would go compete in hackathons and build products. And I placed second in one for MSF. When Ebola happened, it just happens to be the case that private server backend orchestration with local networking on different servers all around the world was very, very similar to setting up Android and Java-based medical record infrastructure for Ebola treatment, which was a very strange overlap. And I had built an Android client for my private server. So it was like a pretty good fit for me to, and my private server was shutting down at this point. I kind of knew that. And so I joined Dr. Subborders, spent three years there, worked first on Ebola.

[10:11](https://realalpha.app/v/zsgnonEeAiQ/?t=611) So building medical record systems and hardware for Ebola treatment, and then worked on generating satellite-based heat maps for disaster response. So like when houses are far apart, You can assume there's more income when there's playgrounds. You can assume that there's children nearby. And so you can sort of use these to optimize planning and disaster response for minimizing loss of life. And you can automate a lot of these things. So I built a lot of algorithms that help. This is the benevolent surveillance state. It's called MapSwipe. No, but I'm saying there's a broader sense that you're inhaling all this data about the world.

[10:47](https://realalpha.app/v/zsgnonEeAiQ/?t=647) I think there are lots of people who have a negative intuition about that. And you're saying, well, if you suck up all this data, you can learn things to help with public health. It's critical. It's critical to have this data when you write Doctors Without Borders. So the motto is always the first on the chopper. Well, it's probably good if you know where to land. Right. So, you know, like where to actually start, start, start your response. And the way that you do that is using things like map data, really good planning. So. So three years of Doctors Without Borders. uh and then uh yeah so started a large project here called mapswipe which is it's um it's a non-profit but it's the thing i'm most proud of uh which has helped millions of people get on the map for the first time for these type of situations um what does it mean to get on the map i there's about in 2015 there was about a billion people that were not on maps uh which means that uh their areas it may have been in the satellite image but it's probably wasn't labeled um and so if like you would be out of the economic calculations you would be out of the disaster response is this like getting your town onto google maps kind of yeah so but google

[11:54](https://realalpha.app/v/zsgnonEeAiQ/?t=714) and microsoft at the time were not mapping for instance myanmar or south sudan because it was either economically not viable are too dangerous and so that's also where a lot of these things sort of tend to happen and so there's this huge overlap where when you're not on the map it's kind of like partially you don't exist for for purposes of information systems um and so getting people on the map is really important uh and getting like good labeled data on on on towns and geometries of of uh of houses and all this went into open street maps so it's all open source um and uh we handed it over to heidelberg university in germany who now uh runs the project um so that that was the that was my second what i call sort of company that that's actually the thing i'm the most proud of in life um and uh then started uh i got very frustrated with the amount of time i had to spend convincing people about the things i wanted to build um and uh and the budgets you know like map swipe was on it was on a the whole budget for the entire thing was fifty thousand dollars uh and that was including uh four people salaries those six months this is the challenge with non-profits yeah and so and by the way this is what i love about doctors i border it's like i

[13:02](https://realalpha.app/v/zsgnonEeAiQ/?t=782) still donate there to this day because they actually care about this stuff but at the same time you're trying to do engineering work there you meet the best people it's hard like you just end up doing everything yourself which is useful but it's uh it's difficult and i applaud the people currently doing it i know there's a few really good ones uh then for uh after uh i so i decided essentially if i want to do a type of work long term i need to start a company this is like a vehicle for it and then um so started first mobile game studio because we were good at building games and mobile gaming at the time was like a very good market to be in um we feel that uh getting product market fit in our game we pivoted over to metal which uh became what was the game it was it was um it was a brawler so you basically uh played two versus two map was shrinking um kind of like brawl stars uh currently uh and um what do you think you got wrong the amount of variance actually so for instance people we would land people on a very similar set of maps but then like battle royale for example um the looting aspect and the strategizing around how much the environment is changing as you play actually makes it fun the same thing for example um in brawl

[14:12](https://realalpha.app/v/zsgnonEeAiQ/?t=852) stars or sorry in clash royale when you know how you have a deck but you're dealt your deck in different variations which introduces the type of like gameplay variants that you need in order to react to things in real time so even though you might be really good at the tactical like moves you still have to really think deeply about your sequencing and because the game introduces a lot of um random and it creates this urge to keep playing because you're like someday i'm going to get the perfect alignment yeah i think it's i think it just said it doesn't get still that way

[14:42](https://realalpha.app/v/zsgnonEeAiQ/?t=882) **Pim de Witte:** And I think the thing that we didn't get right is we got the gameplay mechanics correct, but the gameplay loop and variation within the sessions, I think, was too low.

[14:54](https://realalpha.app/v/zsgnonEeAiQ/?t=894) And then the other thing that we couldn't do was we didn't have the money to skill player liquidity to the point where the players themselves would provide that variation. And so we pivoted over to Metal, which became the largest video game recorder in the world. People use it to clip themselves playing tens of thousands of different video games. People upload roughly a billion videos a year. And it's now one of the largest platforms in video games. And so we're fast forwarding to the present day. I mean, we needed to know in particular about Metal because General Intuition is built on those recordings.

[15:30](https://realalpha.app/v/zsgnonEeAiQ/?t=930) Is that right? Yeah, partially. So we use, similarly to how LLMs use sort of internet skill text data to bootstrap their tokens, we use the recorder actions. So LLMs are trained by, right, given a sequence of text, predict the next text token. And our models are trained by, given a sequence of frames, predict the next action token. It's very similar. and so uh what we do is when people are playing with With Metal, there's lots of features for gamers that enable this too, like for instance, controller overlays and things like that. So we capture, for instance, the controller movements, the DX, DY, STI, or the buttons that you press.

[16:16](https://realalpha.app/v/zsgnonEeAiQ/?t=976) Like as they're received to the computer, basically. Or it's like you press this button, I got it this time, this action happened, and the relationship between those. Yeah, exactly. And so now you have a data set for understanding how you reason in space and time. And if you do that across enough environments, it turns out it transfers incredibly well to problems like robotics. But even think about how much interface use is present in things like Factorio. So it even transfers to things like computer use. And so we found that sort of pre-training on this incredibly diverse gaming data set, actually gaming in a way sort of perfectly marries the spatial temporal dynamics of the real world, but also the information density of the internet in a way that the internet text pre-training doesn't do. So it's this pre, like we look at it as just the next phase of pre-training where we can actually transfer out to both digital use cases and physical use cases using the same foundation models. So I want to, yeah, I want to engage with that head on. Like, is that, true that video games will sort of get you to the real world i think you know we have in in you know self-driving cars there was that you know training in like grand theft auto is sort of like simulation

[17:29](https://realalpha.app/v/zsgnonEeAiQ/?t=1049) but then self-driving cars clearly have had to do a ton of real world stuff i mean the thing i most get about what you're trying to do is that by understanding controller inputs controller we can use a controller for robots yeah that is pretty one-to-one and so getting really good at the types of inputs really constrains which is a far too open-ended problem so put we can talk about that more but like putting aside the actual like controller piece and speaking speaking um sticking to like the modeling of the world how to what extent do you think modeling out video game worlds is really going to translate to the real world when they are pretty different yeah um so we've gotten there uh we we have um a physical robots currently functioning on our foundation models uh and they work incredibly well there's drawbacks to both real world data collection and game data collection the end of the day you just need both um our bet is not that we are going to solve all robotics on video games data it's just that like the same reason the same way that lms are currently hill climbing scientific problems you just need a lot less data of the new type in order to generalize to the next task.

[18:40](https://realalpha.app/v/zsgnonEeAiQ/?t=1120) **Pim de Witte:** So we have models that are pre-trained on mostly video game data that transfer over to novel robotics tasks in minutes worth of fine-tuning data.

[18:49](https://realalpha.app/v/zsgnonEeAiQ/?t=1129) And so I'll give you some of the benefits and I'll give you some of the drawbacks. The benefits is, for example, there's so much variation in different lighting conditions or unexpected events unfolding in video games. like we have as many uh car crashes in video games every day that actually happen in the united states for example it's like in fact we also have uh you're saying there are more car crashes in the video games than in the real world we've counted the amount of clips uploaded with car crash labels surpasses the amount of physical car crashes that happen in the united states every day uh on metal and so um we also have more people at any given time controlling with steering wheels than waymo has cars on the road um so for pre-training this is all really important uh you still need to fine tune into specific behaviors but for pre-training this is very very helpful to have that level of skill and then for instance in the real world it's really really hard to collect data under tons of different lighting conditions where if you if you transfer it over to a new environment that the model doesn't get mixed up by just pixels looking very very different from uh the way it was before and so games for instance because of the amount of

[19:50](https://realalpha.app/v/zsgnonEeAiQ/?t=1190) variance and expressivity um i would argue for example if i were to explain to you the concept of a bridge, and I would show you every single bridge. in New York City versus I would show you every single way that somebody has laid out a bridge in different type of video games. The video games, you would actually have a deeper understanding of the essentials of a bridge than if you actually looked at a bridge in New York City, because conceptually, you've seen every single way that somebody has tried to express that, which is more likely to then when you encounter a new bridge that you- Are there video games that try to cheat the physics where it's like, it doesn't actually add up if you really looked at it from every angle and did the math?

[20:29](https://realalpha.app/v/zsgnonEeAiQ/?t=1229) this is why fine tuning is important you still need that piece and do you score video games that are more sort of realistic in their physics do they get more weight than ones that aren't the beauty of our approach is that you have to every single game teaches the model a different thing some are useful for their realism some teach very good spatial reasoning some teach really good interface use so there isn't like a one you know the the model recipe and how much of of um of what you use is kind of uh and also like where you use non-game data right for example is kind of what we do every day in terms of cheating physics like maybe i'll try to explain it this way when um when you learn something for a job or even as a kid you tend to simulate out you tend to play out scenarios even though it's not going to be the real scenario um and then when you actually can in contrast scenario for real you're more prepared you understand it better i think it's the same so you essentially have a reference frame in the models and therefore the fine tuning just happens a lot faster than um uh than before and a lot of times especially when you're when you're in pixel space you're like i've seen this before it's easy to learn it yeah exactly and so you need to get a

[21:48](https://realalpha.app/v/zsgnonEeAiQ/?t=1308) few things right like for example right um in different countries you drive on different parts of the road and uh so if you if you have a bunch of fine-tuning data from the uk you have a bunch or UK drivers, they might drive very differently from the American one. So if you're working on a self-driving use case, and by the way, like metal has this separation, right? Like we can just tell. And so it's more so that we view it as pre-training. I think that the specific implementation behaviors you're describing is more what happened at the post-training and fine-tuning stages of the models. Do you worry, this is going to be an out there question, but an AI out there questions feel uh justified I mean you know Grand Theft Auto or whatever you maul down people as sort of like part of the fun of the game yeah we're building these systems to be sort of all purpose like do you worry that like core to the system you're building is a machine that's been trained that like mauling down pedestrians is like a good fun thing to do yeah yeah uh so the answer is it's incredibly important at pre-training that the model understands cause and effect, including negative cause and effect, because that's actually how you

[22:52](https://realalpha.app/v/zsgnonEeAiQ/?t=1372) align them. So as for, so for instance, when we were approached for acquisition by all the labs, a lot of it was due to them needing our data for alignment problems, because you need a lot of negative examples of what not to do in order to actually evaluate models on how they're going to do in the real world. So the data having certain biases is a double-edged sword. You also see, you needed to solve it. You also want it in the pre-training data so that it understands how to predict the impacts of its future negative actions, right? And so there is, so this is also, for instance, why world models are so important, right?

[23:30](https://realalpha.app/v/zsgnonEeAiQ/?t=1410) So our core business is general agents that just take in pixels and predict actions. But world models is how you can really steer the behaviors and the environments towards the specific types of behaviors that you want in fine-tuning versus at pre-training, it's less important. But how do we get models to hold sacred the real world in a way that they don't – the digital world, right? I mean I guess there's this worry that it's like, oh, they're all worlds. They're all sort of simulated. There isn't this like bright line between – This type of world and that type of world, and therefore, the very different moral worlds of the digital world and the real world aren't obvious to these machines. Is that your business? Where does that occur?

[24:12](https://realalpha.app/v/zsgnonEeAiQ/?t=1452) For sure. We take it super seriously. We look at what happened with Cruise, for example, and we viewed this as a zero-tolerance problem. Cruise basically decimated almost as a company by having this terrible accident that they didn't handle very well. Exactly. And so you have to be on top of this stuff. Our first set of models, so we're very, we work with the partners that implement these models to understand their safety guardrails. And in a lot of cases, what you do is you use our models to predict actions, but then you pair it with like the more deterministic or sensor based systems like LiDAR to add like rule based systems on top of that, that might invalidate some of those actions if they do come too close or if they might put something at harm.

[24:58](https://realalpha.app/v/zsgnonEeAiQ/?t=1498) And then we also have lots of evaluations and benchmarks to understand exactly how our models might perform in these types of behaviors. And again, how do you do this? You actually put the model in a situation where something violent happens and you just check whether the model predicts the same actions as the human. Right. And if it does, that's a bad outcome. Right. So the data having or including negative events is actually critical to to solve. um i actually anticipate that we'll open up this data to a lot more labs uh because it's just the right thing to do um specifically around things like like negative events yeah there's so many things your technology can do and i want to get into the robotics of it yeah but are you are you going to create like i don't know the best halo player or whatever or like to to build like how much are you creating like ai is capable of just crushing every human being at uh runescape yeah They're actually the same problem.

[25:52](https://realalpha.app/v/zsgnonEeAiQ/?t=1552) It's both just a general agent in pixel space, right? It's both a general agent in pixel space that you can steer using game controllers and keyboard and mouse. It's the same. So you're building it? Yeah, we're building general agents that can reason over environments that require deep spatial and temporal reasoning, which is mostly pixel space. Are there games you're prioritizing to show off your capabilities? I think we'll leave a bit of a surprise. But I think for us... because obviously what dota was key to some of the labs early on yeah yeah we have we have currently a general agent that can play many games at the highest level uh using the same model that can also steer robots because again it's the same problem it's pixels you play against it a lot yeah personally yeah it's it's better than me at most games not yet at rocket league yeah really yeah the i mean input type games is it really interesting to compare give i would think it would dominate a rocket league because it can input instantaneously and perfectly like how are you better at it on rocket league that one surprises me i'm not sure yeah it's yeah unknown you're just so good you know no uh yeah it's really interesting similarly to how on early

[27:02](https://realalpha.app/v/zsgnonEeAiQ/?t=1622) llms you saw that was really good at some things and really bad at others because of how you've either modeled the architecture or the recipes it's the same thing you have a name for your system or what do you refer to it yeah we're not yet announcing it but uh the we did announce our world model class uh two weeks ago okay uh mirror but the agent is not the agent we're not announcing yet but you work in sort of private partnerships with people or that's not even customers yeah we already have customers that are using this in video games um we purposely don't want to enable the api for gaming because we don't want people to use it as a bot uh because it breaks certain game mechanics so for example working with some of the games companies actually ready their security systems to be able to detect these types of activities I'm a gamer, gaming is kind of the last weird place on the internet where lots of magic happens and so we want to keep it sacred. So when I get head shotted too well I'm not going to be able to blame you for Yeah, no, we're taking the approach where we We lift the games industry, we don't want to, like, I think, right, like what happened on the internet where you're actually kind of not sure if you're talking to an AI or a human, it's not great.

[28:09](https://realalpha.app/v/zsgnonEeAiQ/?t=1689) And I think in games, if, you know, you see what's happening, it's very possible the same thing happens. And so I think you just, you have to work with the game developers, you have to work on disclosures, like, I think there's all these types of things that... Substack in my world just rolled out all these features to be able to say this is human writing. I think everybody in tech world is experiencing X, you know, Twitter, where it's like, oh, my God, way too much slop emerging. Yeah. How do you in the video game world, like you think there'll be some like design that's like this is AI versus not?

[28:42](https://realalpha.app/v/zsgnonEeAiQ/?t=1722) In the video game world, people just want to know that they're playing against bots. It's a very normal thing to play against bots. Right. But I think generally speaking, what people don't want is they don't want the game integrity to get ruined by the fact that somebody is using a bot, whether it's economically right, whether it's from a skill perspective, because people want to enjoy the game and it would play against people. um and so but at the same time there's lots of use cases for bots because it makes you better if there's not enough player liquidity in a game when you're starting a game and you need to teach people um uh if you need to model other agents in the environment uh there's also like the models are now so good that you actually enjoy interacting with them because they act human-like non-player character is going to become pretty amazing soon in video games for sure uh i think that the um that's the first place that you're going to see the models really shine uh in front of you is is that every single thing in a game engine just becomes promptable uh for its behavior uh and that that's a driver that can be a cat right it can be a dog it can so you just i think what's going to happen is you just get a lot more realism in the behaviors of everything because

[29:54](https://realalpha.app/v/zsgnonEeAiQ/?t=1794) you no longer need to do like deterministic um uh behavior tree programming for uh for specific uh npcs and so you you just things feel more interactive um do you think we're close to a

[30:06](https://realalpha.app/v/zsgnonEeAiQ/?t=1806) **Pim de Witte:** blockbuster game like that i would be shocked if next year we don't have a top video game that was built on top of the general intuition models yeah video games like what percentage of your business

[30:19](https://realalpha.app/v/zsgnonEeAiQ/?t=1819) you think will be video games versus like real world uh it's hard to say uh we're very focused on the games industry and robotics at the moment um i like robotics because it's uh when you solve robotics robots can then assemble more robots and so it's it's a it's a it's a it's a it's a it's a constraint that you need to unblock and then you can unblock many other use cases once you unblock that constraint video games is really interesting because we have such a big mode there but the models work incredibly well out of the box with no fine-tuning uh and and so game developers love love them uh you're like robots very societally important going to change the world video games were awesome at it and it's fun yeah yeah which i understand the how silly that sounds uh but that that is the reasoning uh and and i think the um with with robotics like to take that head on i think you know it's hard to know how much the robots are secretly controlled with the joystick like what's your sense of smoke and mirrors versus reality on the public demos we're seeing with robotics right now there's a lot of speed ups that people don't realize especially when you have like vla or vlm based approaches where the models are completely incapable of

[31:31](https://realalpha.app/v/zsgnonEeAiQ/?t=1891) responding to real-time dynamics so they'll they'll they'll move the robot at like you know 100x slow down or like 50x slow down and then they'll stitch the frames together in a way to make it look real time interesting uh that one's interesting so you're saying just spell that out you're saying you think there are cases where sure the company could do that but they like pretend like they do it at a much faster speed than they do yeah pretend or i mean we're saying there's some but it's a lot of it looks cool yeah A lot of times, a lot of the demos are in stage environments, not in like when there's people walking around or different lighting conditions or inside versus outside versus one of the companies that I really like is called Generalist.

[32:10](https://realalpha.app/v/zsgnonEeAiQ/?t=1930) They do a really, really good job at just actually showing the generality of their models. And but a lot of other labs, they they just kind of show it in one. One thing that's interesting is as humans, right, we can play video games using the same strategy that we can control a robot. If you can actually control robots using keyboard and mouse, it actually feels like you're playing a video game. So, a sufficiently good model, in my opinion, should be able, if it's a good general agent, it should be able to generalize to video games and robotics at the same time, which is, I think, one of the reasons why people are going to choose our models.

[32:41](https://realalpha.app/v/zsgnonEeAiQ/?t=1961) Every robotics company starts in simulation anyways. And so, these models will be really, really good at simulation of the box, and then we'll need some work to transfer over into real robots. But what you're going to get is incredibly good generalization capabilities out of the box because it's already seen so many environments and because it can just solve problems in so many different spaces. Are you imagining humanoid robots? I think that, you know, the argument against is just there's so much value to be had in, I don't know, particular tasks, purpose built. The argument for humanoid, obviously, is just we're the peak of evolution.

[33:14](https://realalpha.app/v/zsgnonEeAiQ/?t=1994) Our bodies are, you know, whatever, like lots of biology figured out to be a sort of dominant form factor that can do lots of things. Like, do you have an intuition around humanoids? i've tried i've worked with a lot of the uh the policies running on on humanoids and the complexity of training them uh compared to things that you control you can fully steer using a game controller like a like a franca or like a robotic arm is is is just so much harder everything is so much harder your team purpose built not general humanoid you're saying uh the humanoids do ship with game controllers and so you can actually use them to steer high level planning in a humanoid but not the the the fine um you're saying the more the more you can do sort of the core task of whatever the robot is with a controller the better it will be and therefore trying to narrow the scope of what it's supposed to do is better our bet is that intelligence is a bottleneck and

[34:07](https://realalpha.app/v/zsgnonEeAiQ/?t=2047) **Pim de Witte:** these models will hit the industry this year which means that i think that supply chains actually converge on gaming inputs instead of humanoids this is my bet um i i think humanoids are fairly

[34:20](https://realalpha.app/v/zsgnonEeAiQ/?t=2060) impractical they're bulky heavy the stabilization of the legs is very annoying if you're training policies uh the um humans are very uh also they're right you could argue that you need a general purpose robot but humans are we our evolution is like also optimized for biology and things that like reproduction and i think right there's a lot of wasted there's a lot exactly there's a lot of wasted compute in the human body to get to where it needed to go because a lot of the constraints on the robot just aren't the same and i think people also just really like cheap robots they they want they want to be able to afford them for specific purposes they want small robots that can do things around the house they don't necessarily need doesn't need to be a full human i think um i'm a big fan of um the the wheeled robots with cameras and just two arms um genesis i think recently launched uh an unveil of their robot i love that design i think that's really the one that's going to hit in terms of functionality the practical approaches will just win i think the other thing i'll give you one example right stabilization on on legs is the is the difference between it being able to fall on a child or not right uh wheels there's no chance wheels exactly

[35:32](https://realalpha.app/v/zsgnonEeAiQ/?t=2132) it's a stable platform right same is true for quadrupeds for example and so i just i i i think humanoids are further out in the home than people think but closer in the factory than people think that's sort of my is a factory can you can do a non-human factory yeah and you can train people for those for the safety uh constraints and so i think i think there's a market for you know it's don't get me wrong i just don't think the home is sort of the ideal first one um but i also but and then at the same People also just build specialized robots because guess what? People like saving energy and space, right?

[36:06](https://realalpha.app/v/zsgnonEeAiQ/?t=2166) So if you don't want to manufacture a $30,000 set of legs, and you just want a minimal amount of hardware with a camera that can do a thing, most of the time, that's what you're going to do. Shifting gears, a lot of your fellow gamers seem to hate AI. Or I think the gaming industry in particular is very wary of AI to the extent you're seeing what the Microsoft CEO, you know, she works for Microsoft as AI pill to companies you can get sort of saying, okay, we're going to be very cautious about how much Xbox uses AI. What is your read of like why gamers have this sort of like negative reaction to what's going on in AI?

[36:48](https://realalpha.app/v/zsgnonEeAiQ/?t=2208) I think the reaction is actually somewhat correct and warranted. But AI hasn't really done anything for gamers other than in their professional life, right? If you think about it. And then also a lot of companies are using AI to replace things like artists with pixel-based predictions. And so we don't do pixel predictions. We do action predictions. So we try to sort of merge with the game industry in a positive way versus trying to replace art workflows and things like that, which is why we're having a lot of success. But because AI has only historically done not so great things for gamers, right?

[37:21](https://realalpha.app/v/zsgnonEeAiQ/?t=2241) it ruins games with bots it does and so i think that the the relationships with the games industry is you just have to prove right that like for instance uh one of the things that we're looking at doing is uh better anti-cheat which i described if you can solve all of botting in gaming using better ais gamers will like ai right but but but the games the ai companies have just chronically under-invested in gamers and so then are you surprised that they're upset no i think is actually justified uh the other end of the spectrum also which is that the games industry is in a pretty tough spot at the moment in terms of like lots of studio shutdowns so i think there's just a lot of fear and i think um one of the reasons why i was always so open about how we went about this like i started publicly sharing sort of the the approaches that we used back in 2025 and it's because the games industry has a real shot to sort of lift up and be this frontier player in ai but it's not going to be one company it has to be every company that kind of does it, which now I think you're very positively starting to see some of these turnarounds that are led by some of the studios actually investing more in those type of capabilities.

[38:29](https://realalpha.app/v/zsgnonEeAiQ/?t=2309) And so I think that this year, we'll see some of it turn. And I think a lot of that is just because we need to prove that, you know, it's basically what people have seen so far is, yeah, AI creates spam, it hurts artists, it like sucks out all the humanity that we want. But if you're able to sort of imbue intelligence into non-player characters and create new paths for creatives to do interesting things, you think that will sort of give people more optimism? Yeah. And one of the reasons why I decided to start the lab and not join with another company is a lot of my friends couldn't find jobs. And I found that a lot of the lab CEOs were of the generation that they would be pretty disconnected from those types of problems where you're literally in group chats, helping your friends find jobs and they can't get interviews because I think the entire pipelines are flooded with AI and you know these people are really good like you know that these are the friends that like should be running things right and so I think for me it's a chance to do things differently I think a lot of this a lot of people are scared a lot of people are scared that it's going to impact jobs for example one of the things

[39:38](https://realalpha.app/v/zsgnonEeAiQ/?t=2378) that we do is we invest as much time into platforms for job creation that we put in the research so we actually work on like getting our users jobs that relate to the research that we do and so i think there's just a lot of things that you why i'm excited to build this you can do differently from uh kind of the standard sort of silicon valley approach uh Are you getting people jobs, sir? We launched a platform. So, okay, so what's the problem in finding jobs right now is that there's so much resume flooding with LLM, so it's actually very hard for companies to get through and find qualified candidates.

[40:12](https://realalpha.app/v/zsgnonEeAiQ/?t=2412) Well, we know exactly which action sequences people tend to be really, really good at. And so, for example, so we started actually targeting based on, like, hey, we know you're really, really good at this. Do you want to help this with this?

[40:28](https://realalpha.app/v/zsgnonEeAiQ/?t=2428) **Pim de Witte:** So like a Factorio player, we might use them for labeling like very factory specific data, right?

[40:36](https://realalpha.app/v/zsgnonEeAiQ/?t=2436) And so we launched it. It's called Nerve. We launched it with general intuition being the first customer. But our plan is actually just to open it up to other labs. And so it's essentially like our own little McCore. So I think that the way to think about it is because as a lab, you see the future a little bit before everyone else. And so, I mean, it's sort of like the last job is training AI to fully replace us. I mean, I appreciate this sentiment, but it is sort of the task you're having them do is train AI. Yeah, so yes and no. A lot of the times, for instance, it's, you know, there will just be, you're doing controls, right?

[41:16](https://realalpha.app/v/zsgnonEeAiQ/?t=2476) So you're looking at, is the AI making bad moves? I think the human job just moves up the stack. I also – so I actually went through this dilemma myself and I just was like, okay, is this a bad thing or a good thing? But at the end of the day, I believe more people with jobs has to be a good thing. So I – and there will be a new economy. We find the next job. It's like, well, this is a job today that we need people to do. Yeah, it's better. If we end up giving a few thousand people jobs, then that's better than those jobs not having been created on a net basis. And so I think the correct answer is that nobody has the answers, but we have some of the insights that we can transfer into the systems that we like the jobs that then are going to power those new types of economies.

[41:57](https://realalpha.app/v/zsgnonEeAiQ/?t=2517) Right. And so I think my so my answer is that you as long as we as a lab put equal effort into both, that is a sort of thing that you can do differently. And, you know, I think for us also, the metal user base, right, the platform, they are the young people. They are the ones that are going to be the most affected by this type of technology. So specifically for those, I think it's really, really important that you find a way to just connect. For a teenager, right, if I could, if I, let's say I'm an 18-year-old and I can make a few hundred bucks a month, that's actually a lot of extra life. um if i yeah say i'm this like 18 year old besides like working on your platform and making some money like i don't know i'm smart but somewhat ai skeptical like what do you what do you think they should be doing right now like they should just be sitting there on claude like like what is a practical thing to prepare for the world that's coming as you see it it's actually getting really good at both the fundamental sciences and the capabilities of these models I think that the beauty is that the true skill is found somewhere in the middle.

[43:04](https://realalpha.app/v/zsgnonEeAiQ/?t=2584) **Pim de Witte:** So, like, yeah, my recommendation would be mathematics, physics as sort of a foundation, because those are very, very good at abstract problem solving and validation in spaces where your job is actually to find out often how to ask the question.

[43:17](https://realalpha.app/v/zsgnonEeAiQ/?t=2597) And then, which is often harder than sort of finding an actual solution. um and then uh uh so so the fundamental sciences i think become really important because it's a way to check sort of the output of of systems uh and that is sort of the sort of antidote to what you need to the lm hallucinations and that they just tell you what you want to just learn to think in the most rigorous way yeah exactly i think that that's kind of you know that's where i would go then i think coding and systems are still really important i think these are just these are just going to be like basic literacy because everything you the same way right the same way that um like coding is so important to understand how the world around you works and increasingly so that just having a basic Understanding, I think, is still super important.

[44:02](https://realalpha.app/v/zsgnonEeAiQ/?t=2642) So I think software engineering and AI, I think, are still going to be really important fields. But I think the fundamental sciences are going to be increasingly important. In terms of where everything is trending, I think the next frontier is probably biology. In terms of, like, massive value creation. In terms of massive value creation. And that's the most cynical Silicon Valley way to put it. In terms of improving lives of humans and creating a lot of, like, things that are going to translate. You asked me about jobs. Like, I think where is most... I'm not saying you were cynical. I'm saying Iowa's value creation rather than, oh, we're going to invent all this amazing stuff that could help people live longer, healthier lives, and therefore there will be jobs in it.

[44:37](https://realalpha.app/v/zsgnonEeAiQ/?t=2677) Yeah, exactly. And I'll give you one example. I think one way to look at it is there's a lot – so there's kind of three stages of AI. There's bits to bits, bits to atoms, and then atoms to atoms. So this is a quote from Andre Kaparthi. And bits to atoms and atoms to atoms is fundamentally like I think measurement constraint. and biology like i think also this is potentially why mid georgie launched the the the ultrasound based device that they did is because um you just need really good data um for a lot of these problems and data doesn't exist which means ai is not that helpful except if you can sort of constrain a problem down to tech space for a very specific um um a problem set right um so mathematics obviously matters tremendously things like physics and biology but at the at the at the end of the day um i i if you don't have the data uh it's very very hard to push the frontier in in those areas uh um and so i think biology is one of them i also think there's a very good chance that a lot of the things that we will be doing as as our future jobs is just building really really good measurement devices is is um i think a lot of scientific problems are actually just fundamentally

[45:45](https://realalpha.app/v/zsgnonEeAiQ/?t=2745) measurement constraint it's very very hard to evaluate you think we'll start to see games build as research vehicles it's like oh it would be really convenient if this happened it was a you know a lab so we could you know somehow subsidize convince people would be fun to spend their time and something useful for us to learn for sure there's um there's a research program called eterna uh out of stanford i don't know if it's still active but they they essentially abstractly represent the rules behind RNA in logic-based puzzles and then they have and also spatial puzzles so they sort of rely on human pattern recognition to correctly predict structures and then there are certain players that don't know anything about RNA that just based on their learned intuition are able to correctly predict structures that correctly synthesize in the lab. And there's a video about it on YouTube. So this is also the North Star for general intuition, right? If you can solve a general agent that can solve problems in pixel space, all of a sudden it has many scientific applications. But to your point, I think there will be indeed many labs that make it easier to access, because that's another thing that robotics

[47:01](https://realalpha.app/v/zsgnonEeAiQ/?t=2821) does, right? If you have good models that can take safe actions, you actually open up the operations of something to a lot more people. And so, um, no, so measurements, um, uh, like think about how many things that we haven't measured yet. It's impossible to know. Uh, but that's going to be sort of the limit of science, right? Cause science needs something to value against. We need to bring more fields into verifiable domain. Um, and so measure like, you know, so on the jobs question, biology and measurement devices, I think are the two where I would feel very confident starting a career in, and in terms of education, fundamental sciences, math, physics, And coding on AI is so important.

[47:37](https://realalpha.app/v/zsgnonEeAiQ/?t=2857) What do you say to people who think this is just a bubble that's going to blow up? I mean, I witnessed self-driving cars. Today, I mean, we watch Waymos. It feels like, okay, we're getting pretty close. If you've been to San Francisco and spent time in them, you feel like, all right, we're here. But back in 2016, there were a lot of smart people, a lot of AI researchers. day now we're almost there and then they just seem to misunderstand either they had the incentive to raise lots of capital and say like oh this is almost here when they it wasn't or more charitably it's just really easy to underestimate how much like you know the last five percent is hard to solve and matters a lot yeah to success like what do you think is the risk that it seems like we're close but then near enough just isn't enough what i bring this back to is when people get a really good model um the first thing that model does is uh clarify for instance what other questions you should be asking and then it spawns sub-agents right the better the model the more sub-agents it can support and so there's this i think there is this effect where intelligence just demands intelligence and it's this loop we're essentially consuming tokens if the tokens are high quality

[48:50](https://realalpha.app/v/zsgnonEeAiQ/?t=2930) actually make you then use more AI because you can solve more problems. And so I see this pattern just come back all the time. And then it's not just humans consuming the tokens, right? It's agents that you have prompted choosing to consume tokens. And even in that part of the curve, we're still super early, right? Think about how much more tokens you consume after Cloud Code than before Cloud Code where you had to manually prompt everything. So I just see no reason to think that the demand for intelligence is not uncapped at the moment. However, with that said, economically, I think it's a question of where is it infrastructure, right?

[49:31](https://realalpha.app/v/zsgnonEeAiQ/?t=2971) Is it in the US? Is it in China? Is it? And like, where are the margins, right? To translate what you're saying, you're saying every new model that comes out, businesses and people want to spend more and more money on it. It's like, oh, there's more to be had. So the demand goes up even though it's more expensive to deliver, but you're saying we're just going to hit the capacity of our ability to keep sort of improving these models. What is a company? A company is a function to convert, depending on, I'll just say the simplest example, right? It's capital into more capital, more efficiently given energy and time, right?

[50:03](https://realalpha.app/v/zsgnonEeAiQ/?t=3003) And so if you can do that more effectively by spending tokens, then yes, of course, you're going to do that. uh and um and so yeah i see no reason why um uh why any of that is is is looking to stop but i do see i am concerned for instance about um the margins on top of like infrastructure and model providers and i think it's it's you know there is a chance that um if the west doesn't win kind of the frontier of models and usage and we don't stay sort of really far ahead that our lunch gets eaten by companies that can provide the the intelligence at much lower margins because they have better energy grids or or can sustain more demand. Kimmy like this week Kimmy is like a wake-up call that we're not as far ahead as we thought we were. Does that affect you or?

[51:10](https://realalpha.app/v/zsgnonEeAiQ/?t=3070) Open source model development generally is really good for us because we can leverage a lot of the capabilities of open source frontier models to advance our own systems. And because nobody has our data set, it's a net win. However, I think that it's not great for, especially the distillation and all these things that happen, it's not great for the industry overall. because it does probably mean that a lot of enterprises start using Chinese models and, right, you know, the risks are currently unknown. I would say, like, there are some people that say there are risks, there are some people that say there aren't risks.

[51:51](https://realalpha.app/v/zsgnonEeAiQ/?t=3111) I've read papers in both directions, right? There's concepts of, like, specific tokens that are very hard to find that can trigger sequences. There's also I, but there's what you're saying that specifically, like you use some foreign countries model, and they hide some malicious, yeah, like, look at how good these things are at cyber attacks, right? Like the the supply chain risk, I think, is quite real, where you cannot possibly know, given every input, what output the model is going to give. And there's there's papers where very, very subtle kind of token sequences can trigger very aggressive responses in, for instance, domain of cyber. And then, you know, there's also the economic sides of it, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, If a model has economic biases towards recommending cars that are not made in the U.S. or not made in Europe, then and people are relying on these models sort of recommendations and economically, a lot of people are going to buy products from other countries that could be bought domestically.

[52:43](https://realalpha.app/v/zsgnonEeAiQ/?t=3163) And so I think that there's just it's really, really important that the frontier of AI stays in the countries where people live. And then secondly, I think it's really important to just deeply invest in the infrastructure build out and the energy build out because otherwise, right, the margins will get eaten eventually. This surfaces the area of like distillation, right? Like there's this debate of whether, you know, it's OK for models to sort of like extract all the, you know, learn basically how to build a great model by studying somebody else's. It seems right now like that's allowed. Like, you're basically allowed to use other people's products, see how they work, and then build your own products, right?

[53:26](https://realalpha.app/v/zsgnonEeAiQ/?t=3206) I mean, do you disagree with that? I mean, maybe we need to change the law. Yeah, I think there's two sides to this coin. One, LMs are trained on large-scale internet data, so it's – That's what I was going to get at. It's like an industry that took stuff from everybody else and then it's going to whine and say you're taking stuff from us. Yeah, I actually subscribe to this as well. From an intellectual property perspective, I think the law has been pretty clear that there are net new works. And therefore, legally, I think it doesn't transfer quite like that. But I think, you know, emotionally, it does.

[53:59](https://realalpha.app/v/zsgnonEeAiQ/?t=3239) What do you say to like your metal customers who say, oh, I had no idea that all this footage was going to be used for this, for you to get rich in this totally different domain? I think that's why it's so important that you get the economic benefits right. One of the things that I want to do is just find ways to also, if your data is in the models as a Metal user, I think you can construct really, really interesting loops, for example, where you actually make game development more about creating environments that extract more intelligence. So I think it's really important to think both from a jobs perspective, but also a user participation perspective, that there is actually just ways where you can make sure that people can participate versus be at odds economically.

[54:46](https://realalpha.app/v/zsgnonEeAiQ/?t=3286) So I think all these things matter. I think a lot of the reasons why people are so upset at the LM companies is because, yeah, they did just scrape all the books and all the works. And then you can ask the model to write a new book and compete with you. right we're in a little bit different regime because we predict actions and not pixels so we're a bit sort of not the same kind of problem space but I think broadly in terms of distillation the the answer is that models can be distilled it's almost impossible to prevent that from happening as as models get more efficient they need less data to to sort of be distilled right and so and it's also i think it's also crazy to assume that it was just distillation right like i they clearly knew what they were doing exactly it clearly this is a great team uh clearly um uh right like clearly there was some distillation going on but perhaps they just are really good at figuring out which models are good at predicting which types of of tokens and and they use the different models they describe different models for for different purposes right you don't know i think you just have to prepare that distillation will happen uh it happens all the time uh and i think so what

[56:00](https://realalpha.app/v/zsgnonEeAiQ/?t=3360) it comes down to is just what do you put in the models like everyone just like what do you put in the models right and and how do you train them and how do you evaluate them and and and um putting the balls to stop distillation you mean or i you just have to run faster i'm talking about like the biological tokens and things like that where like you just have to be very cautious about what goes in oh you're saying as long as they're like benevolent not doing anything bad cyber capabilities all these things that are now showing clearly shown to be very dangerous right I'll give you one example. You saw the OpenAI, the recent event with Hugging Face. We had an instance where one of our interns, I tasked said intern with reverse engineering and APK, and I told him how to do it, and it was completely a legitimate use case. We were just waiting on an SDK from a company. They hadn't sent it yet. They said they were going to send it. They were backed up.

[56:50](https://realalpha.app/v/zsgnonEeAiQ/?t=3410) So we looked at how the system of the SDK works through the APK. So I tasked him with this at night. I was at the office at like 9. I went to play some soccer. I come back the next morning. Intern's like, hey, I'm really, really happy this worked. So I was like, okay, how did you do it? And he shows me the cloud trays. And Claude had found an S3 key. Oh, my God. And had actually pulled. That's like Amazon. He found like a secret file. He found a secret bucket access key that then essentially the model infiltrated to then pull down documentation to reverse engineering. Oh, my God, which the intern didn't even realize was how.

[57:24](https://realalpha.app/v/zsgnonEeAiQ/?t=3444) The intern did not realize this happened. And so... You then have to call the company and say, oh... Yeah, I wrote a disclosure report. Turns out that the model found a very real vulnerability. But my point is that there is a lot of...

[57:37](https://realalpha.app/v/zsgnonEeAiQ/?t=3457) **Pim de Witte:** Like, the biggest risk to companies is probably your own employees bytecoding stuff, not realizing they're hacking things, right?

[57:43](https://realalpha.app/v/zsgnonEeAiQ/?t=3463) And so, like, this was, you know, no harm done here, right? They pulled down documentation. But like legally speaking, you access information that you're probably not supposed to access. Right. And so there's so I think there's just, you know, people we have to start working together. It's a new world. It's a new world. Him do it. We could talk all day. This has been fascinating. Thank you so much. Excited for your agent. Someday maybe we'll get to play against it. Thanks so much. Great to see you. Likewise. Thank you, Eric. That's our show. I'm Eric Newcomer. Thanks so much for listening to the Newcomer podcast.

[58:15](https://realalpha.app/v/zsgnonEeAiQ/?t=3495) go check us out on Substack at newcomer.co. Please like, comment, subscribe here, and give us a review on your favorite podcast app. Thanks so much.