A look at startups like General Intuition working on large action models, aka world models, which are trained on videogames and simulations, to pilot robots
Engineers and investors pile into world models, aka ‘large action models,’ to do for robotics what ChatGPT did for writing and coding
Context & Ripple Effects
General Intuition enters a field already contested by Google DeepMind, Meta and Nvidia, whose race to build world models has centered on learning from video and robotic data. The distinction in this coverage is an emphasis on videogames and simulations as training environments for models that must choose actions, not merely interpret inputs.
The company’s category-building push follows its July 2026 world-model overview and General Intuition’s confirmed $320 million funding round. Separately, Enigma’s emergence to study human-robot interaction points to a complementary need: evaluating how robot behavior works around people, not just in simulated settings.
First-order effects
- General Intuition and comparable startups gain a more legible proposition for investors and robotics engineers: action-oriented models trained in simulation to pilot robots.
- Google DeepMind, Meta and Nvidia face a sharper benchmark for their own physical-world models, with attention shifting from model demonstrations toward whether systems can select and execute robot actions.
Second-order effects
- Robotics developers will have greater incentive to treat simulations and videogame-like environments as training infrastructure, alongside video and real-world robot data.
- Companies developing world models will need to distinguish the reliability of their action policies from broader visual or language-model capabilities as the category attracts more capital and talent.
Third-order effects
- If simulation-trained action models transfer effectively to real machines, robotics software may consolidate around reusable control models rather than task-specific robot programming.
- The pattern points toward a robotics stack in which model builders, simulation providers and real-world evaluation groups such as Enigma occupy distinct but interdependent roles.
The trend: Robotics AI is moving from perception and language assistance toward general-purpose models trained to make physical actions, with simulation becoming a central route to scale training.