YourInfoDaily
General Intuition Raises $220M at $6.2B Valuation to Train AI on Gameplay
Valor Equity Partners and Khosla back the Medal sister company turning billions of game clips into world models.
General Intuition, the AI company training models on video game footage, has raised another $220 million at a $6.2 billion valuation. Valor Equity Partners, Atreides, Seven Seven Six, Point72, Khosla Ventures and General Catalyst backed the round.
The company is led by CEO Pim de Witte and is a sister company to Medal, the gameplay clip platform on track for three billion uploaded videos a year.
Training on Action, Not Just Pixels
Most video models learn from footage alone. General Intuition trains on action-labeled video: clips paired with the inputs a player actually made, which gives a model the link between what it saw and what it did.
That data pipeline is the reason the company exists inside a gaming business. Medal users upload the clips; General Intuition trains foundation models and world models on them.
"Leading foundation models in robotics and world models are trained on less than 1% of the action data we are capable of training on," de Witte said.
The MIRA Model
The company's MIRA algorithm generates synthetic video for training other systems, running at 20 frames per second at 720x576 resolution on a single B200 GPU with just 5.6 billion parameters.
Its selling point is duration. MIRA is designed to run without the drift that degrades most video generators over long sequences, which matters when the output is meant to train an agent through extended tasks.
"We train directly on the largest dataset of these, similarly to how Tesla trained FSD, but across more actions," de Witte said.
Where the Money Goes
General Intuition plans to scale its team in New York and Europe, hiring more AI researchers.
SiliconANGLE described the technology as a stepping stone to physical AI, noting it has so far been demonstrated on game footage with commercial testing underway.
Why It Matters
Robotics companies face a data problem that money alone does not solve: collecting real-world footage of a robot attempting and failing at tasks is slow and expensive. Game footage is abundant, already labeled with player inputs, and full of edge cases.
The open question is transfer. Whether a model that learned from virtual worlds handles physics, friction and failure in a real warehouse is exactly what a $6.2 billion valuation is betting on.
Source
SiliconANGLE — "World model startup General Intuition closes $220M investment"