What happened
General Intuition, a spinout of the gameplay-clip company Medal, raised $320 million at a $2.3 billion valuation to build agentic 'world models' trained on action-labeled gameplay data. The round was led by Khosla Ventures with participation from General Catalyst, Jeff Bezos and Eric Schmidt, bringing total funding to $454 million.
The company trains on hundreds of millions of hours of gameplay with embedded action labels to develop spatial and temporal reasoning. It demoed a single model that both plays games and controls a quadruped robot fine-tuned on just eight minutes of real-world data.
Why it matters
Using gameplay as a training substrate is a bet that games encode rich, transferable reasoning about acting in space and time. A model that generalizes from a game to a physical robot with minimal real data would be a meaningful step for embodied agents.
Most of the funding goes to compute via a CoreWeave deal, and the company plans to broaden API access by the end of summer. Its founders committed to no lethal-autonomy use cases.
MintedBrain take
Cross-domain transfer from games to robots is promising but early, so treat the single-model demo as a direction rather than a finished capability. The API timeline is the practical signal to watch.
Discussion
Sign in to comment. Your account must be at least 1 day old.