This is gonna be fun.

Millions of hours of people playing video games looked like entertainment. It turned out to be the world's largest dataset of humans perceiving an environment and deciding how to act in it. General Intuition saw it first. Great WSJ piece on world models and AI's leap into the physical world.

First sneak peak in the WSJ on how GI’s large action models work!

gradually, then suddenly all at once 📈 on track for 3 billion clips this year.

Starting to realize just how big of a deal this was

I have been playing these world models/games since the first diffusion DOOM, and multiplayer at 20 FPS is pretty neat. "A dream of rocket league," indeed.

Today we’re introducing MIRA, a new multiplayer world model, built with @gen_intuition, in collaboration with Epic Games. We release an in-depth technical report, dataset, as well as an online demo that you can try right now (link below).

Can’t believe how cool this is. So proud of my teammates and so thankful for our partners

The term “world model” is tossed around a lot but this is the real deal. From video and actions, the model learns a consistent representation of the physics, 3D environment, and game state of Rocket League. I'm excited to have been part of this project because I see this as a step towards learning world models that can generalized to the real world. I’m a believer in learning from synthetic data (Sintel, SURREAL, AGORA, BEDLAM, BEDLAM2) and I’ve seen how models trained on synthetic data can generalize. MIRA is an important step because it’s learned from video gameplay generated completely by bots. This provides a pathway to scaling and scaling is key to real-world generality. Traditional graphics and games provide a path to learning rich models of the world. Check out the detailed tech report: mira-wm.com/paper/

Introducing MIRA. A playable, multiplayer world model. A dream of Rocket League. Trained on 10k hours of data collected with publicly available bots, MIRA learns the dynamics of a four-player game. The model runs in real time at 20 fps, based on the keys you and the other players press. Built by General Intuition and @kyutai_labs, in collaboration with Epic Games. Not used to develop Rocket League. ▶️ Play the demo, read the technical report, and explore the open-source code at generalintuition.com/mira At ICML? Find us at Booth 111 to try it yourself and dig into the results with the team.

Come hang out with us!

ICML 2026! 🇰🇷 We're heading to Seoul next week. Come find our research team at Booth 111 to talk world models and action models - the core of what we're building at General Intuition. We're also hosting a party on Wednesday, July 8th for a curated group of researchers working at the frontier. If that's you, the event RSVP is linked on the thread. See you there!

it does seem like utilizing text & video to scale up to physics understanding of the real world is likely not enough. google is doing a bit of this with street view + maps data + simulation for this but the video game unlock seems absolutely legit esp if you have the full end to end loop of consequence mapping for ai to learn from that’s insanely rich (e.g. controller input → camera movement → game state transition → physics/collision response → npc reaction → player choice → reward/failure/social outcome). i personally grew up playing a ton of videogames as a kid including one of the first games to ever have open ended world simulation (it was called shenmue, absolutely pioneering, look it up). these guys seem to have a good understanding that these feedback loops in actual game play is the best data set for ai to behave in human contexts because that’s precisely how we learn & explore when we are young (like watch a kid play around).

.@gen_intuition CEO @PimDeWitte discusses using data from gamers to train AI & broader implications for the labor market: cnb.cx/3QHDIXj

General Intuition CEO @pimdewitte, who's building foundation models trained on video game controller input data ("action-labeled gameplay clips"), says general intelligence won't "taste like an LLM": "We have a scale of data that's going to allow us to jump to the frontier in one capability — which is any system that can be controlled with a game controller (which is most robots) — and then, you can use that to create a sufficiently general intelligence." "As humans, the decision to talk or type is a very, very small subset of the actions that we can actually take." "So in order to create a sufficiently general intelligence to play 10,000+ video games, the model has to be able to predict across the entire action space of human cognition when they're interacting with these environments. Which are 2D and 3D environments, interfaces, long-horizon tasks, short-horizon tasks, [etc.]." "It has to be a sufficiently general intelligence in order to predict actions. Therefore, the type of model you get out is not going to taste like an LLM. This model is going to be incredibly good at navigating unforeseen environments. It's going to be incredibly good at zero-shotting any task that can be done with a game controller."

Appreciate the credit, but it was really your team we were betting on and the approach you have. As you know, it took one meeting and two hours on a Saturday afternoon to commitment on the spot. This was the first time I actually met you. Self driving vehicles, flying drone swarms, robotics, video games, and much more, all from the same model.

Come meet the MTS and MOPS (member of ops staff = me) in Seoul at ICML2026!

Most AI today has learned to see the world. The harder and more valuable problem is teaching it to predict what happens when someone acts in it. @gen_intuition is building the large action foundation models that do exactly that, trained on billions of clips from over 17M monthly active users on Medal. Congrats to the team on their $320M Series A. Large action models are the next big leap, and this category will produce several of the most valuable companies in AI. We're backing General Intuition because it's built to lead them.