World Model

What is a world model?

Language models are great at predicting the next word — that's why they can write and chat. But a lot of the real world can't be handled by guessing words. A car has to know whether it's about to hit a pedestrian. A robot has to know whether the cup will shatter when it lets go. A video model has to know which way a ball bounces.
A world model is what teaches AI how the world behaves — not the next word, but the next state of things.

How is it different from an LLM?

One lives in language, the other in physics
An LLM learns the patterns of text; a world model learns the patterns of space, time and cause. Show it a ball being thrown, and it can infer the trajectory — rather than just saying the words "the ball is flying".
It cares about "why"
A world model isn't just aiming for plausible, it's aiming for correct — physically correct, logically consistent. That's the step from an AI that can talk to one that can act.

Where you already see it

Self-driving cars predicting traffic, embodied robots planning moves in unfamiliar rooms, video generators keeping frames coherent — all of it leans on the same idea.

Bottom line: a world model is AI's "physical intuition". It's how AI steps out of the digital world and learns the rules of the real one.