World Model as Training Gym
General Intuition
This means General Intuition is trying to own the training loop, not the simulated world itself. The world model matters because it gives robotics and game customers a place to run thousands of practice attempts cheaply before using a real robot or shipping a live game feature. What customers ultimately buy is the action model that comes out of that loop, which is closer to a deployable worker than a demo environment.
-
The product workflow is concrete. A developer provides task data, trains or fine tunes inside the learned simulator, then pushes the resulting policy to a robot or game system. The quadruped example, where only eight minutes of real world tuning followed simulator pretraining, shows why the simulator is valuable as a cost reduction step.
-
This is different from companies that try to sell the world model itself as a content or simulation product. General Intuition is closer to a model supplier for behavior. The simulation is the gym, the shipped artifact is the agent that can move, react, and complete tasks in another company’s environment.
-
The broader market is moving in the same direction. NVIDIA positions Cosmos and Omniverse as tools for generating training data, prediction, and policy evaluation for physical AI, which supports the idea that value often sits in the trained robot behavior and the workflow around it, not in the simulator alone.
Going forward, the company’s upside comes from turning world models into a faster and cheaper factory for action models across games and robotics. If that factory keeps cutting real world data needs, General Intuition can become infrastructure for how other developers train embodied AI, without needing to own the robots or the game worlds themselves.