Last updated:
World models: the next state of reality
An LLM predicts the next word. A world model predicts the next state of reality—what happens when you push this object, drive around this corner, perform this surgical cut. It does not just describe the world in language; it simulates it internally.
Not the next word—the next moment in the world
Instead of predicting the next word the way ChatGPT does, world models predict what would happen next in the world, modeling how things move and interact over time. They can also be thought of as digital twins: a digital replica of a place built from real-time data, where simulations can play out to anticipate what may happen next—before metal meets the floor, rubber meets the road, or a tool meets tissue.
Why embodied AI depends on it
Here is why this sits at the center of embodied AI: world models are the brain that makes the robot body work. Without a world model, a robot can react. With one, it can plan.