Definition
World models are learned systems that simulate how an environment evolves in response to actions, generating predicted future states rather than relying on a hand-coded physics or game engine. In generative video-game simulators like MIRA, a world model produces each next frame directly from a neural network's learned sense of game dynamics, letting it improvise plausible worlds it was never explicitly programmed to render.
Episodes covering this
Worth reading next
Papers we haven't done a deep dive on yet, but would recommend on this topic.
- Genie: Generative Interactive Environments
- Diffusion Models Are Real-Time Game Engines
- Revisiting Feature Prediction for Learning Visual Representations from Video (V-JEPA)
- Emu3: Next-Token Prediction is All You Need
- Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Task