World Models: Computing the Uncomputable

General Intuition argues that World Models, trained on video and actions, are a new kind of foundation model that could surpass LLMs for environments requiring deep spatial and temporal reasoning. They predict the next state from the current state and action, which turns costly simulations, such as a stadium of thousands of fans, into a fixed-cost neural network pass. Yann LeCun's AMI recently raised $1.03 billion, and Fei-Fei Li's World Labs has raised more than $1 billion to pursue World Models. NVIDIA's Jim Fan predicts 2026 will be the first year Large World Models lay real foundations for robotics.