
General Intuition Raises $320M to Train AI Agents on Billions of Gameplay Clips
Backed by Khosla, Jeff Bezos and Eric Schmidt, the New York lab is betting that action-labeled video-game footage is the richest dataset for teaching AI to act in the real world.
General Intuition, a New York lab building AI that learns to act by watching people play video games, has raised a $320 million Series A at a $2.3 billion valuation — just three months after a $133.7 million seed round, with Khosla Ventures leading both.
The company trains its models on billions of action-labeled gameplay clips drawn from Medal, a game-clip platform with 17 million monthly active users. The bet: that human decision-making in games — split-second reactions, planning, improvisation — is the richest available dataset for teaching AI to act, not just predict.
The key ingredient: action labels
What makes the data valuable, the founders say, isn't the footage itself but the action labels embedded in it — precise records of exactly which buttons a player pressed and when. That turns raw video into a supervised signal of intent and reaction, the kind of data world models need to learn cause and effect.
Vinod Khosla framed it in terms of the leap that made LLMs work: "In world models, I think the quantum leap is the emergence of intuition in the AI, a human intuition-like capability. The human action and reaction data you have in games is the key part to the emergence of intuition."
Star-studded cap table
The round drew participation from General Catalyst, Jeff Bezos, Eric Schmidt, former F1 champion Nico Rosberg, and researchers from Google DeepMind and MIT. The majority of the new capital is going toward compute, via a deal with CoreWeave.
Why it matters
General Intuition sits in one of AI's hottest frontiers: world models that could power robots, autonomous vehicles and embodied agents. The thesis competes with Yann LeCun's AMI Labs and Fei-Fei Li's World Labs, but takes a distinctive data angle — mining the exhaust of gaming rather than building physics simulators from scratch. If action-labeled gameplay really does teach machines to act, one of AI's biggest datasets may have been hiding in plain sight, inside 17 million gamers' highlight reels.
Newsletter
Get Lanceum in your inbox
Weekly insights on AI and technology in Asia.


