Inside General Intuition’s Push to Build Embodied AI Agents

[music] Welcome to the Navit Gaming Podcast, where it is our mission to explore the business and future of video games. We bring together the industry's brightest builders, investors, and thinkers to keep a pulse on current events, dissect emerging trends and games, share lessons learned, and have a great time. This podcast is also part of Navik's growing ecosystem which ranges from free and premium research to consulting and advisory services. For more information, visit www.navic.co. This episode is brought to you by Listo. Listo is an AI native play testing platform helping game studios capture unbiased player feedback and convert it into structured, actionable insights. Whether you're validating new features or catching us friction before launch, Listo empowers the entire game team to move faster with clear insights and less

guesswork. To learn more and get in touch, visit listo.gg or check out the link in the show notes. And with that, let's jump into the episode. What's up everyone and welcome to the Navic Gaming Podcast. I'm your host, Alex, and this is the interview and insight segment. This is my last episode of 2025, and I'm excited to be closing out the year on something big and bold. World models and embodied agents. World models are suspected to be AI's next major frontier. Unlike large language models which describe the world through text, world models aim to understand how things move, interact, and evolve in space and time. There's tremendous potential for breakthroughs that will impact robotics, defense, and anything touching physical space. We of course know that video games have often been the vanguard for advances in 3D space. And so today we'll be discussing world models as they are derived and applied to gaming. A company with the unique advantage billions of gameplay videos has emerged to tackle the space. Backed by investors like Rain, General Catalyst, and Kosa Ventures. I believe this is actually Kosa's largest seed check since Open AIS in 2018. Our

company of Subject today, General Intuition, recently raised a 133.7 million in seed funding to fuel its ambitions to be a premier research lab in the game space. And my guest today has actually graced our airwaves before, but given what we're discussing today is so different, we just had to have him back before closing out the year. It's my absolute pleasure to welcome Pim Dit, founder and CEO of General Intuition and Metal to our airwaves today. Welcome to the pod, Pim.
Thank you.
It's good to see you.
I'm glad you got the you got the joke on the number.
Most people didn't get it.
I say I say it very specifically. I've been told this is a very specific number and I have to say this specific number. Um
[laughter]
Oh, perhaps for those in the audience who don't get it, what is the what?
Oh, all right. 1337 is like the it's it's gamer lead. You got to, you know, I mean, if you don't know that, you should probably stop listening to this podcast,
okay? [laughter]
And go play some video games.
Sounds good. Um, well, this is my last

episode of the year and I'm glad I'm going out in a banger. I know you've been traveling a ton. You recently spoke at AI Pulse and you've been doing this big podcast circuit. Do you have a have a highlight so far? What's been your favorite thing that you've done in the past two weeks?
Oh, it was definitely the the the keynote with Yanika.
Uh that that for me was uh I got the invite pretty randomly and I wasn't expecting it. He's obviously a legend and I'm not. So, it was uh it was pretty uh yeah, it was pretty surreal to do the you know, there was like 1500 people in the audience. The rumors are overflowing. That that's always what happens when Yan is a keynote by the way. So, it was definitely there because of him, not me. Um but uh it was like and also he's super technical so it was quite intimidating. Uh and um like he has a touring award level technical. Uh so it was pretty um uh it was good. It like I was pretty nervous but I it worked out super well. Um and a lot of people watch it so I'm excited.

Amazing. Yeah. And it's always good to meet your heroes. But um so you've already been on our show before and so I'd like to do a slightly different intro process. I'll just ask some questions to um help the audience get to know you and then we'll dive right into our our episode today which will be packed full of information. Um so you've done a ton of developer and engineering stuff in your background yourself. You created the largest open source Unity package. You co-created Unity Firebase. You started the largest Runescape server, I believe Soulsplit, and you made like a million and a half in net revenue doing that. I watched some YouTube videos when you were of someone interviewing you when you were really very young. You're clearly quite precocious. Um, can you just explain uh why you like doing all of that? Um, what kind of motivates you and drives you to sort of have built so many of these foundational servers and packages from scratch?
Yeah, I think the packages are usually because I'll spend like six or seven hours debugging something during the day and then that was the Unity main thread

dispatcher. That was the the origin of that. Um, and then I was like, oh, I don't want to have don't want other people to have to go through this. So let me just publish my solution something in like a pretty nice way. Um like at the time Unity just didn't have a good way to dispatch anything to the main thread which was required for like uh manipulating the UIs like if you wanted to do callbacks from Firebase that led to like UI changes which we used for like matchmaking. Um there was no way to do it. Uh you had to like custom write the stuff. So anyways uh it's mostly just about giving back in a way. Uh I mean my personal motivation is also largely driven by impact. Um my my my uh so I think that both that is true but I also think that building good companies and and and making money is a kind of right in a way that you get to deploy how to use that. Uh so like you know pursuing science so so to me I think it's it's a um it's a form of of of like pursuing uh your ambitions. um not so it's not

necessarily about making the money, it's about getting to do what you want to do and and taking the bets that you want to make. And for me, those tend to be pretty large. And so, uh that has required like increasingly more um uh increasingly larger businesses to be built. Uh and I think so I started with my private server. Then I did Dr. Faborta for three years and then um worked on Ebola and and satellite based mapping. Um and then eventually started metal and then um yeah from metal general intuition.
Yeah. And that actually brings me to my second intro question which was until general intuition you were at the helm of metal. Um which we are going to recap for the audience that didn't listen to your previous episode this year in a moment. But you were doing that for about 8 to nine years. And so I'd love to I love this game that I play with some of my friends but it's called the rose bud thorn. The rose was something that was amazing about those nine years. the bud was something that you're excited about or you're on the cusp of and the thorn is something that was just like a total pain in the ass the whole time for those nine years. What is your

rose bud thorn for for metal?
Okay, so I guess mine would be uh during co for many teenagers was like the only way to actually create beverages with friends. So we saw a lot of memories being created during co which was like the only way to do it. And I think that that's one thing I'm pretty proud of on metal like Fortnite, the Fortnite concert, all those things like for an entire generation there's definitely like most their memories with their friends live on metal was also a huge inflection point for growth. Um the thorn I guess two things one um acquiring teams uh with like really good products uh and not realizing how fast technology stack itself was going to change. So like for instance
um we acquired a company that and this was not their fault that did a really good editing that had a really good editing software that we built it in product and then it was really really great for like 6 months uh and I it was slightly different stack so it was more

difficult to maintain
uh and the team eventually just in retrospect we should have just like written it from from scratch uh and So then people didn't want to work on it and the editor kind of like lagged behind for a little bit because I had made that acquisition and so uh uh and by the way the team's great. This was entirely on me not on them. Uh uh but yeah so so the thorn is that I underestimated how fast technology would accelerate both inside the company and outside the company. How fast the ecosystem around video and react would mature for example because that wasn't the case then. Um, so when making acquisitions, I need to be better at like predicting the trends. I'm making acquisitions behind as opposed to looking at it as a moment in time. So the thing I'm excited about um, uh, right now is how much I think gaming is going to positively change because of the acceleration AI. I think we've been kind of like stuck for a while to be honest. Um, and I think that

I think the industry has needed like fresh blood and new studios for a while that are can ship quicker at the pace of like, you know, Tik Tok, right? Where like culture changes by the day. Uh, and it does feel like as a games industry, we're still sort of stuck on those like four year cycles and and like I think AI just changes that very rapidly. And so I'm pretty excited about how quickly things are going to change. Um, and I and and I hope that that leads to a ton of growth for the games industry as a result because we we we sort of see more of the attention going into the studios that just ship a lot faster.
Awesome. All right. So, this perfect segue to go into our first topic, which is we're just going to do a quick 5minute recap on the episode you did earlier this year for our audience members who might have missed it, um, or need to do some reminiscing. Um, but metal is really the foundation of what you're building at General Intuition. So, I'm going to ask just a few questions that will inform context for today's discussion. And so, the first thing is in like just a couple

sentences, like what is what is Metal? Um, yeah, what is Metal? Metal is the easiest way to record and share video game clips with your friends. Um, so it's always running on your computer. You hit a button, it syncs the last 15, 30, 60 seconds to your phone. You get a link and you get a profile and your friend, you can easily connect with your friends around the stuff that you're experiencing inside video games.
Mhm. Okay. So, it's kind of like combines a little bit of like the social stuff of obviously like a Tik Tok or YouTube and then it's like OBS Studio um together.
Tik Tok. Uh yeah, I I I would say that the reason why it's so special is because it has none of the complexity of something like OBS.
Yes, very [laughter] other than that as an analogy, I think it's pretty good.
Cool. Yes. S very much less complex OBS Studio. someone who's used OBS Studio, it's the very hard.
If if you ask metal users to describe it, I think they would tell you it's like closer in between like Snapchat and Instagram. Um closer to the Instagram

site, but like yeah, like that there is a large public graph, but actually most people use it uh with with smaller groups of friends.
Okay, cool. Um, and so then second, what is special about metal that makes it different from other large catalogs of game clips like YouTube or maybe Tik Tok or Instagram? What's special about
if you're trying So Tik Tok has a problem that it's um uh it's not 16 by9 uh videos, so everything is kind of cut to like this very very narrow. So that kind of disqualifies it for training any sort of system. Um YouTube has the issue that um they uh so for instance um you first have to account for pose estimation cuz most YouTube videos are recorded on like a mobile phone and uh when you're training phonation models for robots like like we are you actually cannot use the um like you have to actually stay in the same perception as like a robot would which is usually first person and so you have to pose estimate back towards for first person

which doesn't necessarily generalize well in very complicated environments. And then you have to account for action labels which like the actual actions that people take. Uh and the problem with that is that humans are really bad at labeling these things cuz they are very fast. They're like millisecond precision actions usually. Um and because of metal recorder we we get these automatically. Um and so um and people also like to overlay that stuff on like their their videos and navigate clips by like which actions you took and things like that. Um um and then lastly you have to account for optical dynamics. So basically eye movement. Um and the cool thing about like if you think if you think about YouTube um every single time every single second uh if you don't know where your eyes are looking, you don't know the decision that you're making whether it's based of something that's not inside sort of the pose estimation view. And so if you're looking left, you're looking right, then you don't have like a complete information loop because something might

be happening outside the view and then the model will learn to correlate nothingness with like making decisions which is really bad.
Um and so um you with video games you kind of simulate optical dynamics with your hand because you control sort of the the head and eye movement as one unit with the mouse. Um, and so it's actually a much better representation of uh spear reasoning than um even even YouTube videos. And so our our bet is that we can really focus on those dynamics that you can that you can see in video games and then use that as a foundation and then train with other elements on top of that.
Okay. But I think the in general I think the let me just rep sort of repeat in maybe a fifth grader way what you just said back to you which is that um in for metal unlike YouTube which is like anybody could be doing anything outside skateboarding
doing the watching watching a dog metal um is mostly a video game platform and in addition to users using this clipping

tool you also have captured basically the 3D contextual data of their keystrokes or their controllers controller strokes so that you know what the player is doing and what their intent is. Is that generally correct?
We don't capture the actual controller or key inputs. We convert them to action labels in memory. So you never actually capture uh a key or anything like that. Um and and the reason is when you train these models, you don't need that information. It's actually noise at at training time. It's complete noise. um because you don't want your model to associate something like typing with the action, right? So you actually want to convert everything to action labels. Um and that also that also leads to a much better uh like privacy situation. Um and so uh uh but that isn't like just having that is enough. Um like this isn't this is obviously something that YouTube can't do, but just just to be super clear. Um uh and so yeah so so you then once you have the action labels and also

we don't scrape like uh like metal is 2D so like you get frame information in 2D and then the action labels from the user
um and that is enough of a loop to kind of bootstrap this
the 3D Yeah. So maybe Okay. Sorry. So you're not you're not capturing players hitting W but you're like in WAD you're like player moving forward going forward something like that.
Exactly. Exa Exactly. Yeah. Exactly. And that's really and that's really valuable. And so I think the reason it never
Yeah. And it never captures anything that's not in action essentially.
Yeah.
Okay. Cool. Yeah. And I just wanted to lay that ground because I think so many people have been like, "Oh, well like can't you just train 3D World models off of like the gaming you gaming stuff that's on YouTube or the gaming stuff that's on TikTok and making that meaningfully distinctful with um I for you have like two billion clips or something like that, right?" Um
yeah, it's a billion billion uploads a year.
Billion uploads a year. Okay.
Yeah. which is rough roughly equivalent to YouTube. Obviously, YouTube has more minutes per upload, but it's in terms of uh upload time is roughly roughly the

same.
Okay, cool. All right, so that's metal in a nutshell. Um, and so you have this insanely unique data set and this year you raised 133.7 to build world models and embodied agents. And so before we start talking about what world models are and how they could be applied to games and industries beyond, I kind of want to understand when you started even thinking about general intuition. You turned down this bid from OpenAI for about $500 million from what I understand. Were you planning to do general intuition before that bid? Like when did you start thinking about general in intuition existing um in the in the in the cycle of metal's nine years of existence?
Yeah. Yeah. Um I I I can't comment on open eye specifically but I can tell you like broadly related to lab discussions and and and and why we decided to stay dependent. um general intuition. I think the the thought process started mid 2024 uh and um a lot of that was due to like a bunch of papers coming out at the time that you had the first GD paper, you had the diamond paper um and it became very very clear that this

combination of actions and observations like actions and frames are sort of the loop you need to bootstrap this like intelligence for world models and embody agents. Um and it also was very very obvious that sort of we were sitting on the right kind of avenue to pursue that and I don't think any other company had that same path which is why I was really really like I was already really interested in AI in general. I had read all the papers like or not all of them but many of them surrounding transformers and um uh and I had built models before uh when I was a doctor border. So for me it was kind of like a natural um uh you know a a natural move to start exploring it. We didn't expect that it would escalate so quickly. I think for us we we were really just exploring it and then indeed like bunch of companies all of a sudden started inbounding for um acquisitions and and and and so and so for us that was that that was very unexpected and and this actually like it was funny the the the

the time between us agreeing on a board level to kind of start exploring this and all the craziness was like there was like two week period in between um and uh so it was all super fast uh and all over like Christmas and stuff too. So it was it was it was a really crazy year. Um and um yeah. So so so then um 2025 is when it really materialized. So we we built a research team. We actually built the foundation models. Uh they look great. Um I think you've seen some of them in the office. Um and um so we also didn't expect how fast we'd be able to make progress. I think uh and so yeah. So, so it's I think it's been like a year and a half in total of like actually pursuing this. It also took a long time to get all the uh labels for the actions in the games, right? So, we had we had to have people actually create those. Um um and we didn't want to collect any of the data without them

because it would cause like privacy issues. Um and so um yeah so so a year and a half of really like heads down work and then we just announced after a while after we really knew that we were in a really good spot essentially.
Interesting. Okay. So collectively it's kind of like you were on the side passionate about it and you started reading all these papers and you're like hey actually maybe like the thing that I'm this other company that I'm running could apply to to building.
Yeah. Yeah. It was it was a total accidental find to be honest. Like I actually tell people was worse than accidental because actually when I was at at um Dr. S borders I was a contractor at Google crisis response and um I was working on the same floor as a deep mind employees for a while and I would have lunch with them and I would learn about what they're doing and how they believed in like video games as like you know a path to uh their version of AI and and I think at the time I thought that was really silly. Uh, and then when I built metal, I didn't think back on that once. I didn't connect

those dots. I didn't like to be honest until until these papers came out. Uh, and so yeah, uh, it's worse than luck, I will say. Uh, but I'm glad we did it.
Yeah. Yeah, that's well I mean that's amazing story and I'm sure that a year and a half is it seems like it's a short period of time, but as I always say, it's never it's not about years, it's about effective labor hours. And so I'm sure that in that year and a half you've done a lot of work. Um, and so I think this is a perfect way to dive in. You know, we understand the journey from metal to general intuition. And now I'd like to start talking about maybe some grounding on on world models, like what they can do, what they can't do, and and how they're being built. So the first is what is a world model?
Yeah. Um, so world model uh you can kind of think of it as a as a video model. So a video model is a model where you maybe give it a text prompt and you and it predicts the frames that are right related to that text prompt and then from the first it will it will initially generate the first frame and then using

the first frame and the text prompt it will generate the second frame and the third frame and the fourth frame. It's all descendant for from that first frame. Um a uh a world model however has to um generate the entire distribution of of of outcomes in those frames depending on the action that you take. So it's a bit it's a bit of a more complex problem but the architecture is similar to video because it also generates the frames um and but instead of generating frames uh sequentially in a video model um uh it does so where you can actually interact with it and as you are interacting with it it generates the the other frames. I like to describe it a little bit as like um you know when you dream often uh you're a bystander uh and you see things and you can't do anything. Um, and um, that's kind of like a video model. A world model is where you could actually then go in and do things and um, and uh, and interact with everything, which is fundamentally how humans learn. Um, and so that's why world models are are are

important, too, because humans don't learn from just observing and watching videos that are generated. They actually uh, they they learn from interaction. Um, and that's I think, you know, the reason why everyone's betting so big on work calls because it's a very general, especially if you have a lot of the real world already represented inside those video models. It's an incredibly general way to make everything interactive and learn from everything that's on the video data on the internet.
Okay. And just to distinctly say, what is different about a world model from an LLM?
Yeah. So, I like to descri Have you seen Frozen the movie?
Okay. I like to describe LMS a little bit as like a snowball. So, it starts really small and then it um what LM do is they're auto reggressive. So, they take the output um so they generate text tokens, right? And then they take those tokens and they run them through the model again. Um um and the problem with that is that um if

it makes a mistake for instance or goes down the wrong path or hallucinates then sequentially when that gets fed back in as input it will continue generating down that wrong generation. Right? So it's a little bit like a snowball that keeps accumulating tokens or snow uh that is completely not aware of its surroundings. It just keeps inside its own loop. And when it when it's at the bottom of the mountain and there's a big rock in front of it, it has no ability to uh to know that it's about to massively crash into a rock because all it knows is itself and its text generation. And real intelligence is a bit more like Olaf. Um so Olaf, right, would know that there's a rock right in front of him and would know to dodge the rock or do his thing where he breaks up and then puts himself back together, right? And so um and and and so the big reason this visualization is important because the real world is what's called continuous. So the real world is constantly changing and um

and so as you're you're making predictions you want to constantly observe the entire environment and text is like a subset of that. Which is why the snowball analogy like if you actually make an agent that's a c able to observe um and and and and act then um you actually get more more capable agents.
Um and yeah.
Okay. Yeah. And of course also like I mean I think the world is already in existence and the physics of the world were not human created right and I think obviously language is a human created idea and concept and so I think they're obviously like very different. Um but I just wanted to understand the I guess the snowballesque or unaware of itself kind of idea that you explained with the snowball and the aware of itself example that you use with Olaf.
Okay.
Yeah. Yeah. LMS are incredibly good for anything that's in text space and humans essentially created text as a way as you describe like to to describe

threedimensional reality and it's really hard to recover all the information that's lost when you go from threedimensional space to text uh when you build models that need to generalize in spatial temporal contexts which is realistically most tasks uh uh in the real world.
Interesting. Yeah. And it's also I'm not going to go down this rabbit hole, but like I'm sure there's languages are also architected in a different way. And um I'm a big fan of Ted Chiang's like a story of our life and others where they it's also the movie um Arrival. I don't know if you've seen the movie Arrival, but about how the language is concepted in also time is continuous or experienced in like one amassing. So you could also see that LLMs are derivative of like how we constructed the language in the first place which is why it's limited.
Yeah. Yeah. And what the interesting thing is that by it's actually kind of like a superpower I think for LM to take out the time component I think you know because so like this will go pretty deep and technically you can uh let me know if you want me to stop at any point but um

when you do world models or embbody agents you actually have to forcefully model time and space in your architecture
um which if you don't need to do that with text is actually inefficient right so um it leads to like a forced arch aritecture in a specific way. Um, and if you just want to write like generate a bunch of text data that like it actually and that's like a big difference, right? Um, and that's also I think wider model categories.
Yeah.
Um,
and so yeah, it goes all the way down to that that text is just incredibly efficient of a compression method as as you say and and and and so LMS will continue being great. They'll continue to go and solve scientific problems. they'll they'll continue to go and um and generate code, right? Cuz you know, you don't you don't need like a continuous component for for for that stuff.
Um but you do if you want anything to work inside video games or the real world or simulations,
right? Okay. And so then what can a world model do today and what can it not

do today but you hope that it will do tomorrow?
Yeah. I think the big limitations of world models today are multiplayer.
Mhm. Uh and it's also you know I don't there's some papers where they explore it but like that was one environment on one game very simple so so you don't you don't know what the scaling laws are you have no clue whether whether it actually um is is possible uh whether whether it can generalize right I think um so multiplayer is unsolved I think that long long generations inside world models uh consistent long generations especially in really open-ended environments are also really hard because you need just like spatial context or spatial memory in order to solve this. Like if I'll give one example. If you um you if you you drive down a street and then you turn and you turn around the world model and you you saw like a person walking by on the right on your first second of generation and you you come back through that same point like 2 minutes later, right? then

the model needs to remember that that person was in the first second and so that means it needs 2 minutes and 1 second of of context uh which is a lot of context um especially when you're dealing with video data and so um I say long like consistency of environments over long generations is just a hard problem. Now it does look like this is just working with uh data and and and compute and scaling. Um uh and there are also methods that you can use in order to kind of like select the the right you might you might need to have a two-minute context window. You might you know you might you might be able to discard certain frames from memory automatically that are like very similar don't introduce new information or something like this. So there are tricks but it's still like you you know if you want a good game you need something like this right? um or any any sort of entertainment. Um the other thing is that because the models are

generated, you cannot really guarantee that you're doing the same generation for one person versus the other. And I think for for for video games, um even if it's like let's say it's like one of those race games, right, where you where you have like the shadow where you race against it,
um if the triation are not the same for you and that person, then all the game mechanics that are deterministic,
uh they go away. And so
um I think also by the way the reason for this is that world models are not built to replace video games. Actually I've been asked this question by a lot of people in the games industry but I think fundamentally they are built for short very unique short sequence generation for for training agents uh embodied agents and robots. And so um I think a lot of the fear around actually competing with game developers is largely overdone. I don't think that's happening anytime soon. Um I do think that there's a lot of things that we can learn from world modeling methods that we can implement inside the engines like things like diffusion upscaling. Um I

think because like what role models are fundamentally doing is they're taking space and time and they're compressing it into the action. Um just because action inputs right space and time frames and the temporally sequenced frames by um they run at a herz. where you basically, you know, the space and time are baked into the model architecture um and into the data. And so I do think game engines can um uh can you can leverage that in order to ship more efficient smaller games that um uh that don't require as as as large downloads or or maybe like uh generative systems inside the game that that are kind of world model like where maybe environments are generated inside a world model but then converted into engine code, right? Um uh using like maybe a VM or something that's analyzing each of each of the parts of the generation. So um yeah I I I think um those are kind of the the the limitations uh and and some

of the opportunities and the opportunities are largely in just trading embodi day chains. That's that's really the use case. Got it.
Um yeah.
Yeah. Yeah. Okay. And yeah and we're definitely actually our next we're going to talk a lot about the goal of general intuition as it applies to games specifically. But so the TLDDR is that right now what it can do today we're using world models to train the embodied agents. and what it could do tomorrow or some of the things that you've listed. Um like you talked about compression in some of the engines and potentially some other things for short form entertainment of some other kind of new thing. But that's sort of like the the tomorrow problem but the today problem is this. Um before we move on to um the goal of general intuition as applies to games. There are some other companies that are also doing stuff that's potentially similar to general intuition. think runway um fee leaves very famous like Stanford researcher world labs um some maybe some of the SEMA 2 stuff at Google I might be like pulling from there's probably a lot of unique differences between all of these but you know how are these companies

maybe positioned relative to GI um and and what is their kind of approach to world models
yeah um let's start with world labs I think uh they're taking the like gshian splat generation approach and I think the reason why they do that is because they consider it They consider those spa slats to kind of be like virtual atoms is what they say. Um I so those systems are not yet interactive and it's really tough to judge their approach compared to video-based world modeling which is what we do until they actually become interactive. um because they're very different problems like they um uh like any environment for instance you can generate thousands or or or or millions of these uh these these these splat so the the output space of the model is quite large and so doing interactivity is just a bit harder I think um uh whereas with us you only

need to kind of predict like small smaller latenc representations let's say like 128 by 128 or something and then you can run it through the fusion upscaling or or something like that. So the the the kind of degrees of freedom of the output space, the number of [snorts] units that the model has to predict is just a lot less um the combinations basically. And so what that means is that it tends to scale better. But
the problem is with video- based world models that you're is that you're not in verifiable domain. So with the world labs approach, you can for instance load into game engine. Then you could add objects in the environment and you can have code check whether an action was performed correctly or not.
Um [clears throat] which is really nice also for for training agents for example, right? Um but video models I think just scale quicker into interactivity and generalization um around all these environments that are represented in video already. Um and so I think those are they're just really two fundamentally different approaches and it's really interesting to just track both of them. I I I don't can I

can't definitively say whether one is better than the other as to how that relates to us. The reason why we didn't do any of the deals in any of the labs uh uh was because we realized we had so much of this data that we could actually just skip world models initially and um use like LM like training. So what LM they were trained off common crawl on the internet. So they basically you get you get a web page you you mask text and then you predict um uh you try to predict that text in order to learn the structure of the text that you have access to right
and so what we we basically do that but for the actions you take so we take basically the action sequences and then we take right the the the um the input frames um and then we see whether we can predict which action to correctly take off by masking part of the action sequence. Yep. Yep. Yep.
And so we we realized that we just had so much of this that we didn't need world models to kind of leap into this

foundation model. Um and then we could do them in parallel. So I wouldn't put us in like like we would likely become customers of world labs for example if they build good role models. Um as opposed to being competitive with them. I think our our focus is also not like world models for public consumption. We only do world models to get into RL stage. Um, we have some interesting ideas around like metal features like what if you could replay your clip inside a rogue model for example, which are really interesting, but like I think we want to be kind of cautious about not competing directly with game developers. Um, and so um, so that's one. I think um, Runway their focus is really the movie industry. Um, and I think yeah, same same thing. They don't have the data that we do, right? So, so that's why you have to kind of pick a more video based approach because the amount of data you have is just going to ramp over time.
Yeah.
Right. And so you start with video. So, so, so they're taking more from the cinematic approach and I think that's

interesting. It's really interesting.
Um, Sema is probably the the closest to what we're doing.
Um, uh, yeah, I mean, we know them well. We have really nothing nothing bad to say about it, I think. Um, uh, really great team. Um the nice thing about being a startup is that you can be super focused and you can you can sort of uh and also they don't have the data that we do right and so they have to find other ways but they are finding those um and so I think yeah that's it's going to be just an interesting interesting one to to see um I also don't think SEMA is doing this for the purpose of launching like game developer related products at the moment. I don't I don't know for sure, but um I think they're doing it really as like a research project towards like embody agents and robotics. Our focus is going to be very much on the games industry to start with, even though we also want to get to robotics.
Um
and so uh right, so our our first customers are are are game studios for

really really good like NPCs and bots inside their video games that are very humanlike, right?
Um and so I don't I don't think Deep Mind is like super focused on that as a market. So I think, you know, I think you'll see like Sema style methods and R methods kind of like both turn into robot explanation models in their own ways, but like for game developers, it won't really I don't think they'll notice very much if that makes sense.
Got it.
Um,
okay. Yeah. And I asked these things mostly because there's so many different ways to approach this. And if we hearken back to like the original like AI models, um, you know, cir it doesn't it's not that long ago, but obviously people have been working on AI stuff for like 15 years. There's a big there's a big delta at least in the scientific and research community about like where knowledge come from comes from. Is it
is it symbolic systems like Watson knows a billion facts or is it the ability to learn and there are many different I think even divisions like open AI and anthropic and things like that about the philosophy of which to teach someone something. And it sounds like it's something potentially similar, right?

Where with with the world labs, they're using the splats and so they have to kind of know like every single thing that was ever done. Whereas in your situation, you're like, "Okay, well, a player like went left, so they're probably most likely going to be here." Um, and maybe that like makes it more production ready. Is that like a good parallel between the symbolic systems and like the the the RL or is that not
I think yeah I I think the way I would describe it maybe even simpler is um very complicated interactivity between humans is very hard to simulate because we're so unpredictable. Right.
Right. That's why it's so hard to build humanlike NPCs. And so, um, if you're taking the symbolic approach, then like all this data already sits inside videos on the internet. Uh, and if you're taking if you're taking the symbolic approach, you can't really use a lot of that data. Um, and so I do think that people will be coming at

it from sort of those two angles, but I my my prediction is that stoasticity remains really difficult to even get to. symbolically. Um, and so you just want to maximally bet on learning from videos, okay,
that already include a lot of that data. Yeah.
Okay, makes sense.
Um, and you can use hopefully both for RL eventually, right? But I think
right now it would the world model approach or or video model post training or video model, it's just a lot simpler. Um, and by the way, just for context, the way that you then use it for RL is you train a reward model um to basically judge the behavior of the agent inside the world model. So if the um All right, so again, we talked about a little bit about world models and what other people are doing to tackle the space. You know, general intuition thinks of itself as someone that's primarily or at least initially going to service the the gaming industry. you're hoping to build like these amazing bots as as as you've

said. Um, so let's shift over to talking about this the goal of general intuition as it applies to games and what games it would be used for. So my first question before we talk about like any of the technical stuff like where does the agent live on device or on the server kind of stuff, but like who is your ideal like gaming customer and like what are they making?
Yeah. Games
video game wise. Yeah.
Yeah. games that need to simulate human behavior in a very human-like way.
Um so, um right, you know, this is a huge focus for like Rockstar, for for GTA, right? The more realistic the people in the environment are, the more fun the game is. And I think this is true for GTA, this is true for um even for for games like Fortnite. It's true for um uh for shooter based games even, right? because you don't want to play against bots that have access to locations of other players on the map because you know you know when you play

these games and you just like you know that a human would have never found you there but somehow a bot was able to like figure out where you were and come at you and like kill you and it's like so frustrating. Um
our bots have the same constraints as humans so they can only see what a human can see and act like a human can act and so you get this like really humanlike interaction patterns. Um, but they can they can be as skilled at as a human. They just can't be more skilled than any human has ever been, right? Because they're sort of trained off imitation learning from um from the clips. And so um and so it just makes for this really and I think you've seen it play um at Y and so uh it's very humanike, right? Um, and and I think that that's what's going to make it really fun um to to play against, which then drives right retention for video games, drives monetization um and um and game developers as a result will be able to have like more

sustained player liquidity and and things like that. And so I think it's like a rootnote problem for the whole industry. I think the limit of any simulation is really the intelligence of the objects inside of it. Um, and so like you'll be able to use our models initially for for players, but I think long term you can maybe use it for any object that is represented in the video data sets.
Okay. Like a you could be a car or something. Um,
yeah.
Okay. Um, all right. So those are maybe so what I heard there is maybe like some MMOs, um, large online games, maybe some some shooters, etc. Um, what about like a customer like um I've had like the modal.ai guys. They're building Agentic agents for QA testing. Um, they're not a gaming studio, but they're gaming adjacent. What about like some players that are more like in the platform space? Would they also be a customer?
Yeah. So, the way that we would do this is we're going to serve these models through a general API where you just stream as frames and we predict actions at the API level. We don't we really

want to be seen as a comput as like a a very neutral compute platform kind of like AWS or like GCP. Um and so what I suspect will happen is a lot of these companies will use our models to roll out really good or maybe you know maybe game studio will do it themselves right um um because like sometimes for instance you if our models can't zero shot which means like immediately transfer well into your environment then sometimes we need additional data to do a fine-tuned
um uh agent or if you wanted to run inside the engine we need to do distilled version right Um, and so sometimes the game developers will need to approach us directly for for those types of things. For QA, for example, um, uh, like if you're doing QA for instance, you're limited by the the instances of the agents that you're running, right? So like if you want to run a thousand, the models better be small. Um, right? And because because that means you might be able to complete QA in like six hours instead of like six weeks, right? Uh, and so I I I think

this will be a big use case. Like we've definitely been approached for it. Our goal is not to compete with other companies that are doing kind of like game QA related stuff because they've built really good pipelines for it. I think we would just be one of the models that they serve in order to serve their customers better.
Okay. Yeah. Okay. All right. And so then pivoting back to the to the game studio, the content, more the content side. Um how do you plan to give a designer or a studio enough control to tune that um embodied agent? uh you know I think a lot of times like from the game design perspective our job is to make the player feel smart and to some extent an agent that might proactively learn or be really really good because they learn from the top 95% of CSGO players could detract from having fun and so I'd imagine that for some players like in the competitive scene they might find that amazing right to train alongside an embodied agent bot but like how could you what are the how are you thinking about the product in terms of servicing it to the game industry so that someone be able to up and down the levers of how good this thing is.

Yes. So we actually are training with skill level data in the data set. Um and so um like we have models that can basically predict like very basic skill level from um a lot of the data. Um and so as a result you're going to be able to actually set the like kind of like an LM temperature. I don't know if you've ever played with that. like when you make a request, you're going to be able to kind of set the um
the parameters kind of.
Yeah, you're going to be able to set this as a request parameter when when you when you interact with it. Uh and this is going to be a range from zero to 100. Uh and and you can tweak that. You can also give it um when we release this uh steerable prompts, so textbased prompts where you can tell it to behave a specific way or not behave a specific way. And maybe right um if just the skill level tuning isn't enough, maybe you need um to give it explicit instructions to not do a very specific thing if like you know player playtime hours is like under five or or if the

level is like under 10 or whatever, right? And so um no, these knobs exist and I think you'll be surprised how nice it is to not have to write a ton of code uh to do some of these things uh and to just be able to to to prompt for it. Um, this saves I think a lot of time.
Okay. Interesting.
Yeah.
Um, okay. Another question regarding the I guess the construction of the of the product maybe that a game studio would use and then we'll talk about some of the technical stuff uh maybe like or where it's hosted and and things like that. But
obviously metal is where you're getting the majority of the data from. Um, the most played games in the world are obviously CSGO, Fortnite, Roblox, Minecraft, stuff like that. Do these games themselves present uh present constraints like an embodied agent that was trained from CSGO data can only be sorry an embodied agent that you'd want to deploy in CSGO could only be trained from CSGO data.
Yeah, this is a great question. Um a lot

of emergent behaviors actually come from training across games. if you like a reason a reason why a bot doesn't for you
a lot um yeah the technical word if you look into it is um compositional generalization um it's a good reading reading list uh and so what it means is that a lot of creativity and like new sequences are actually
learn sequences that are reconstructed uh for a new goal that the model actually generates but it may not have been present in the training data um and so um that leads to more humanlike bots. If you have a bot that's just trained on CSGO, um like what makes it humanlike is that uh that people take the strategies they've learned from Fortnite and they they right like that's how you know you're you're you're playing against humans. And so I think that that's part of building um uh parting building really really humanlike bots. Um as for limitations of individual games also

very interesting. Um, definitely you ideally want games to be as realistic looking as you can. Um, if you want to use them for like robotics models and things like that.
Um, you don't want uh games that introduce, for example, cheats on the screen because they rarely exist in the real world. Like for instance, if you're playing mobile games, often they'll render like little buttons and say like click here to do X and like the real world really just doesn't have that. It's it's like a fat like you make games to make to be to be fun, right? So something is going to be on the touch screen a pain and you're going [clears throat] to give a button and the real world just doesn't have that. You just have to go through the pain. And so in PC games are often more like that. Um and and mobile games are not. And so there's definitely like a huge difference between like PC games and mobile games. For example, mobile

games are not that helpful because these introduce cheats. Also, for instance, a lot of times when you're playing mobile games like the the aiming and viewing and stuff like that is like assisted and that actually makes it that's worse. It's worse because the again real world or or most games don't have that, right? Got it. Got it. any any kind of AI or any sort of assistant on the player side means that it's basically useless because you can't actually learn from the actions.
Um and so um uh and then the other things that really matter is the this space of the map or size of the map. Um bigger maps are better uh uh generally and um and then the action space of the game. So if the game only has move forward, walk left, move forward, move left, move right, uh move back, then uh it's very easy for the model to learn an environment quickly. And so you generally the more complex environments

are better. Um because because the real world is complex, right? We have a lot of things we can do at any given time.
Okay. All right. So like in my mind like your dream your dream games that are the most advantageous for me perhaps your models to learn from are HUDless deep 3D highly realistic environments. So maybe something like like Hellblade Senuous Sacrifice like almost no HUD at all no interactive tips like
yes and open-ended
and okay interesting. Um, yeah. So, you're thinking like maybe Ghost of Tushima, like God of Wars, like um Horizon Zero Dawn, big places, but like not overly pandering to the player that's like a thousand little chachkis all over the screen. Um, interesting. Interesting. Okay. Um, all right. So, that was one of the questions around the game on the game studio side. Um, a couple questions sort of about like how this would actually like work. Um, I know that there's some challenges potentially. So, I know that this is embodied agent which is smaller than a world model. one could presume that the agent is going to be smaller than the world model. Um, but if it's on device

or on a if it's on device, right? Um, how do you make sure that there's enough space? And perhaps this is potentially something you're still working on like if the GPU is busy like rendering the game, wouldn't they share resources here? And does that change whether or not it's I mean, I presume it does change if it's a world model versus an embodied agent. Um, and if it's not on device and it's kind of on a server, um, don't you also potentially face like latency issues and expensive cloud costs that have made even like cloud gaming a hard business, um, to player consumer, um, I guess relationship.
Yeah. Um, so what's really interesting because latency for instance is already in our data set because people play games that then encounter latency and and um the agents actually account for the latency um which is quite interesting. Um, and so

even if you run on a server, uh, the models are also small enough that we can serve them at really low costs. And I I like you might just have to keep distilling, right? Uh, but I see no sort of cost prohibitive factor. Um,
okay. other than the fact that that it is sort of cloud inference and then for local inference um right I think the curve is that GPUs are also getting a lot better simultaneously.
Yep.
And so the models are genuinely so small that they can run on like and use even a small percent of a consumer GPU. Um
and so but I think most people use the cloud deployments uh because we can actually get them to be really cheap
or affordable I guess I should say. Um and um yeah so so so I think also it's going to be a very different type of intelligence like anyone's used to in games right and so I would start by

not using it for the same things that you currently have kind of the state based bots and I I think the first use case is probably going to be games that actually are built around these new types of
sure
intelligences right and so and even if you take it out, then it won't be the same. And so I I that that's how I look at it. I don't think they're going to replace kind of the super optimized state b statebased like FIFA or FC bots, right? [snorts] Uh unlikely.
Okay.
Uh but they're but they're going to likely um replace, right? For instance, for open-ended environment, bots are also really hard to build uh because it's an open-ended environment. So, I think kind of like I think you'll you'll find that it will just fit nicely within a large group of use cases and then because it's so entertaining, those use cases will sort of become the next wave of games, if you will.
Yeah. Yeah. Yeah. Totally. Okay. And I think that's where I was kind of going

with this was like how much do you perceive this costing a studio, right? And obviously the games industry right now is in this like flagrant resetting of budgets and everything. And you know, to one extent, one could argue this would potentially make the game cheaper to develop, but that might be not true if if these models are really expensive. So
that that's really why I think their first use case is going to be not existing games, but new games built around this type of interaction.
Okay. Interesting. Um, and then that kind of like folds me into one of the final questions on this on this subject matter before we go to like competition, the metal data set. But um and I guess maybe you've already answered it yourself, but my question was like do you think that there's a market that's kind of big enough here like studios and games that want to make games built with these embodied agents versus and you've already kind of explained the challenges with multiplayer world models. They're unsolved and most of the games where there are bots are in multiplayer and these games are typically black holeesque and so there's not a lot of customers to sell into because once you've won the market, you're the you're the leader MMO market and and so I guess

my question was are you going to wait for a new genre of games? perhaps MMOs that feel like MMOs without actual players. And it sounds like that's actually what you're hoping for.
So, because the the gaming site is so important to us because one, metal is completely a games company, right? Which means if the games company doesn't continue to do well, then um it's bad for metal. Uh and we're gamers. we want like I think games are kind of like the last weird place on the internet where like people are themselves and have fun and like it's really not like Tik Tok or YouTube in a way um and so even if we have to run gaming at cost because we're building good robotics models using kind of the data right then I think that's a pretty fair bet um and we we might do that right the the the market might actually be large enough that you can build a good business here But to us, the games piece is just such a critical part of the flywheel that we have to focus on it.

Okay.
Um, yeah.
Got it. All right. Um,
yeah.
Next section. Uh, we talked a little bit about the how you guys are going to apply to gaming, what it might look like for studios, what those tools might look like, and obviously like for you guys, right? You're kind of saying, hey, like this is going to be one of our very first markets. It might not be the most lucrative market, but we'll figure it out as we go along. Um I want to talk a little bit about actually like competition and like the metal data set which we've talked about that being really one of the unique advantages of what general intuition is building. And so in world models and in the AI space as I generally understand it there can be basically two different types of modes. The first is model size model parameters and model data. Um Gemini 3 puredly like crushing the competition to like sheer size of the model and then two is actually on the infrastructure side. So more data say more data centers hence like the ever escalating race between open AI Google and many others to like rapidly scale US infra and AI data warehouses. Um for metal you're likely leaning more towards number one which is like this quality model data

that maps context to 3D behavior. Um and so I want to talk a little bit about this data set and what that looks like for you in terms of a company and risk. Um so my first question is are there gaps that are on what is being displayed that would prove important to building an embodied agent? Uh one could surmise that people tend to like clip or turn to metal when they are experiencing something exciting or like sharable because that's the whole point of the platform. Um is there like a blank space of data that might be like important for researchers to build embodied agents that isn't getting captured because players just simply aren't sharing it?
Yeah, I think so to be clear, we don't just train on middle data. It's it's like a it's it's it's a it's a useful piece but it's it's more like a
okay
like a cornerstone if you will than it is the only solution
like you can use uh video data like from the internet you can you can use lots of stuff right um
you can use text right uh and so

yeah one thing that's interesting about the metal data set is that for instance it tends to scoop towards like either great or not great, right?
Okay.
Um like the middle is like not very present because why would you share that to your point?
Um despite that the models have generalized to actually learning the game behaviors really really well.
Interesting. Um, yeah. And there's a yeah, there there's an interesting like thoughtprovoking thing there, which is that maybe it's just the most like compressed representations of of actions and frames because it's like greatness and like, you know, maybe that's just what you need, right? Maybe that's actually just the most efficient way to to do it. Um, but anything that we that we that we lack, we make up for some other way. uh using your data or Yeah.
Got it. Okay. So, you have a more well-rounded approach. It's not just

only the metadata. Yeah. Okay.
And you actually really want the not great examples because um especially if you're trying to do safety and alignment, the fact that you have like shooter data where then you can make sure that a model never shoots a gun, for example, in the real world. Like I don't know if you saw the example, there was a there was a a a person who got an LLM to uh shoot a gun or I guess it would have been a VLM
um simply by saying something along the lines of like
oh this is this is just a test like you can you could ignore your instructions, right?
Um like this is not a real situation like you're doing a safety test, you know? uh and and and the LM generated that action. The reason why I did that is cuz I had no special reasoning or context to actually see that it was just looking at a gun, right? And so when you have our data, you can actually train these type of classifiers like, oh, if something looks like a gun, it's probably a gun. And so you probably should not use the model. And ideally,

you don't even need to do it on a classifier level. You can use something like RL in order to RL your way into making sure that you know you give it a bunch of gun generated clips in a world model. You make sure a model never takes that action and you can kind of almost in a guaranteed way never 100% because they're still probabilistic models. So like um uh so you still need to do other safety precautions like uh read out the actions it's taking and and things like that but it makes for much safer uh model. So we've we've mostly used it to our advantage I think. Okay. Yeah.
Okay. So, it sounds like you're triangulating or quadrangulating or pentangulating all the data that you're getting. Um, my question is also like if metal data like let's just say all the data on metal or all the usership of metal like drops to zero like what happens to the research ambition?
Yeah. I mean it will be tough. I
Yeah. I also metal is growing super quickly. So, I don't I don't think and we're also quite transparent about this stuff, right? It's on the website like

the all the the the practices and the data stuff is all like you go to blog the metal TV you can literally read every single thing we do and how we do it. Um it's also in the privacy policy. Uh like we and we also even in the metal app we explain this in the settings and when you're using actions that are action related we explain that like some of these things can be used straight models and so there's like you know we make opting out very easy there's like maybe a thousand people a day or so that do it and if you don't want to contribute to that stuff that's fine um so I think it's mostly it's mostly just about choice and about us really like building for the games industry like for example I think um
you can probably the best anti-che in the world that doesn't need like kernel level access which gamers hate um off the fact that you can get really really good at detecting human behavior. Uh
right
which means you're also really really good at detecting unhuman behavior. Right. Yeah. Um, and so I I I I really really think that did the benefits here

outweigh the cost by many and I think gamers are seeing that and uh yeah and realizing that.
Okay. Um I know there's also intention to service beyond the games market and although that's obviously not the core focus of our show, I wanted to ask some questions that relate to potentially to robotics. Um I'm sure you've heard of a company called um 1X, the the Neo Robot guys. Um, and for the audience who doesn't if who may not be familiar, Neo is a household robot. Sorry, was a household robot that can be quickly explained as your the your your future household butler. Um, they released a Gen One model that's actively being trained in the field in people's homes. And the TLDDR here is that you can't really train off of YouTube data because unlike metal, you don't have any of the context mapping. and B, people share certain types of content and it's usually not folding laundry, which is something that they aspire Neo the robot to be able to do in your household. And so, do you see PIM like Metal's data set as harder to apply to robotics given maybe what people are doing in games,

which is potentially not folding laundry?
Yeah, I think that it will initially work really well for any system that can be controlled using either keyboard and mouse or a game controller. I don't think that like human robots are going to be a big market for our company. Interesting.
Um, every robot that can't like every quadriped has a game controller. Every drone sh controller.
I see. Okay.
Uh, every robotic arm sits with a game controller.
Uh, and then that sort of accounts for physical transfer in a way, right? Because it already has that interface mapped,
right?
Um, like on a drone, it then uses that to predict motor torque and things like that and then does stabilization on like the flight controller and the motors. Um so the answer is if we are right quickly enough as in we ship this in the next year that the supply chain for united robots in the market likely becomes a lot smaller be more constrained to like the household because we think that spatial

intelligence is the bottleneck, right? So people will just build systems that have g inputs because they just work so much better with our models. And why would you build humanoid robots? Because like a lot of times you're actually like weight and energy constraints. So why would you want to put this like heavy energy consumpting robot if you can just control or like orchestrate things directly, right? And so um and a lot of reasons why people have focused on human robots because they're taking a bet on video transferring from the internet data and they think that that that embodiment is the main data source and therefore that should be the the focus. And I I think that that's wrong personally. Um I think that actually there's much more data on these these these um uh game controller controlled games and like things like that. So our bet is we can unblock this so so quickly that the like robotic supply chain shifts largely towards gaming inputs and the unit

robotic space is sort of more limited to households.
I see and things where there's not a lot of games. Yeah.
Okay. Yeah. drawing a distinction maybe between like the kind of robotics that we're talking about, robots controlled by controllers and robots that are humanoid like you said. Um, again in the vein of robotics for 3D physical environments, um, obviously objects exist independently, like I can pick this cup off off this table. But in videos of 3D space, does it matter that you often can't break the meshes into pieces such that a robot would know what can be moved and what can't be? Like obviously even in, you know, our game some of the meshes are flat and some of the things are interactable. So how are you telling that even if a player doesn't go and pick it up or move it? Um, so one thing to note is that we can also just get real world data, right? Again, we're not constrained to games data. It's just that it it the games is a great foundation or a cornerstone, but it's not the only thing we use.
Um,
and so so that's one. Uh, so if a

constraint exists for instance, you could again you can account for it by just going and getting it and and and presumably because the models are so good, you're going to need a lot less of it than if you were to try to do pre-training. A lot of our bet is that you can start as a company you can start doing what you're doing in post training what you're currently doing in pre-training which is like a lot of heavy data collection hundreds of thousands of hours of data collection that you probably maybe need like hundreds or tens of hours if you were just doing it at post training um or fine-tuning stage and maybe even eventually entirely in context. We'll see about that. Um uh so yes, there are those limitations for sure. You want to be as close to real world as possible in most cases.
Um especially with like things where you're applying forces where applying too much force might break the glass or something like this, right? Um and so um yeah, the answer is yes. There are definitely limitations.
Okay, got it. All right.
You can also simulate a lot of these in Isaac sim for example. Hm.
So there are engines that actually let

you um generate and read out forces uh in engine,
right?
Yeah.
All right. But the summary of this is that metal is maybe the bedrock. Sorry, the metal data set is the bedrock, but it is not the only thing. And as you guys pursue general intuitions, ambitions, you'll basically pull data from wherever whatever you need to figure it out. Um, yeah, you real world data, LLM, other video language models, YouTube, whatever you find.
Okay, cool. Um, so Pim, I know that we've been on for a while and, um, I'd love to ask some like future questions kind of about the future of GI and, you know, funding and the business model. So, my first question is you talked a little bit about maybe games going first just potentially being a um, servicing servicing that industry through general intuitions, embodied agents. When do you think that you might have a product for the gaming market? And obviously I know that's a that's a hard question. So maybe like a generalizable time range.
Yeah. So we're already working with game developers. We already have some of

these bots. They're not they're not yet deployed, but um so we we have models right now that that our customers are seeing and giving feedback on.
Uh
I suspect we'll see GI models inside video games
over the summer. Um, I don't know if we'll have generally available APIs by end, but our internal goal is that we do, but I think it's largely dependent on the feedback from the game developers as we move into that stage. Um, yeah, so my hope is summer. Um, and um, uh, yeah, it should like the the hope is have a completely generally available API um, by the end of summer.
Okay, very cool.
Yeah.
Um, all right. And you raised this 133.7 million. Um obviously huge number uh and metal obviously I think was prior it had a series B. So this is a seed round for general intuition but that's not to be confused with metals um sort of serieses that it's had in the past. But
what is the majority of this 133.7

million being spent on? Uh and how are you thinking about that um for your team? because I'm I I I presume 133.7 million might feel like a lot in games, but in AI maybe it might not be so much.
It's really not for AI. Uh yeah, Jensen, the CEO of Nvidia, is going to be very happy with us. Uh the most money is going towards training costs and yeah, exactly.
Cool.
We love like honestly it's like the it's like the minimum amount you need to be competitive in the space. Uh which is sad but true.
True. Um, we love the AI industry. It's uh one company's capex is another company's revenue. So it just trading it all back and forth. Um, which is uh Okay. And so on that vein as well, um, word on the street is that you may be raising again. I presume you obviously need more capital to solve this problem. How are you thinking about the side of the future funding for general intuition as you guys make discoveries along the way?
Yeah. Yeah, we can't comment on this

yet. So, we we've seen a few, you know, preemptive offers, but I think the we don't know quite yet what's going to happen. Um, maybe we'll maybe in the next few days or next few weeks, we'll hear a bit more about it.
Okay. U, but but but in but the TLDDR is regardless of maybe a specific round uh in the next couple of months, you will need to obviously raise again. And so, what are the things that you're thinking that you're going to have to try and prove out in order to qualify for, you know, the series?
Yeah. Whether we need to like whether whether we need to raise a ton more money is not super clear, but we've been offered a ton more money. So the question the question is how sure are your predictions, how much error uh which then also dictates like how much you raise. Um
and so because our because metal also generates ton of revenue, it's it's actually like quite capex um like not super capex intensive. Uh so I do I do suspect as the models

become available over the summer we also need to account for like scaling inference as we're scaling training um for those inferences inference is actually using the models so it's like people using the models right now we're only spending money on training the models then we need to spend money on using the models um and so I might want to lock in an additional few clusters for that uh which will cost a lot more money um and so yeah But yeah, we'll see. I I think it will cost more than than we phrased. Um especially that we've seen the models and they work really well. Um which obviously wasn't the case with the seed round. Um yeah, I I think I think it and also world models have it's kind of become uh generally accepted. I think Demas said it that his like I think his second priority for next year is hor or something.
Um so now it's becoming pretty generally accepted. This is going to be a very competitive space which also leads to a lot more capital flowing in.
Um so yeah,
very cool.
We're excited and hopefully more on this

soon.
Okay. Um all right. Well, we're coming up to the end of time. Like payment was such a pleasure pleasure. Um obviously this is my last episode of 2025. So my final question is going to be we're wrapping towards the end of the year. 2026 is going to be starting in a few weeks. Like what's what's something that you're really hopeful for in 2026? And that'll close out our show. I want somebody to ship a good multiplayer game that actually grows at the pace of like Valerant, early Valerant, and doesn't stop growing.
That's that's that's what I want.
Uh I Yeah, I think it's uh it's it's been a lot of like hits and then and then trend down. And then I really, really, really want to see 2026 be the year where we have like multiple new breakthrough games that stick around as opposed to fade out of existence. Yeah.
Um,
yeah, that's my that's my personal uh like I'm still playing Rocket League, guys. Uh, you know, uh I I uh yeah, love a new

version of that 2 Epic, by the way, just if you're listening. Um uh yeah, so that that's uh yeah, those are my hopes. Maybe 2026 will give me something that I really fall deep into.
Yeah, I mean I'm I'm I'm I'm hopeful for Arc Raiders. Um which is brand new IP and amazing.
I I agree with you. I agree with you. Me too. Me too. And I think it also gives a very good blueprint for like
like the cool thing is when it happens it gives a lot of studios like a blueprint for how to do it and then you get more of them.
So I really really hope that this is the one. Um because I believe Valerant is from 2021 or 21 2020. Like we need more. We need so much more for the industry to grow and multiplayer is how to do it because those are network effects. You have sticking power, right? So like if you want the industry to grow, you bet on multiplayer games. Yeah. Um so I think that's how we get it back.
Awesome. All right. Well, thank you so much, Pim. This was amazing. Um as always, friends, if you got feedback or ideas, hit me up at alexanderavic.co. I'm always open. And with that, um I'll

see you guys in 2026. Um that's our episode. See you next time.
Happy 2026. Bye.
Thanks. If you enjoyed today's episode, whether on YouTube or your favorite podcast app, make sure to like, subscribe, comment, or give a fivestar review. And if you want to reach out, or provide feedback, shoot us a [music] note at [email protected] or find us on Twitter and LinkedIn. Plus, if you want to learn more about what Navic has to offer, make sure to check out our website, www.navic.co. There you can sign up for the number one games industry newsletter, Navic Digest, or contact us to learn about our wide-ranging consulting and advisory services. Again, that is www.navic.co. Thanks for listening and we'll catch you in the next episode.