AI scientists race to build 'world models' beyond chatbots

What happened

Prominent AI scientists including Fei-Fei Li and Yann LeCun have launched startups focused on 'world models'—AI systems designed to understand space, time, and physical interaction rather than just text. Louis Castricato left his doctoral studies at Brown University to start Overworld, which is building interactive virtual worlds; Li founded World Labs in San Francisco; and LeCun started Advanced Machine Intelligence Labs in Paris after leaving Meta.

Why it matters

Current AI chatbots learn by predicting the next word in text, but this approach has limits for physical tasks—a robot cannot learn to pick up a coffee mug from a book. World models could enable 'physical AI' and robotics by teaching systems how objects respond to force, how light falls on surfaces, and how an AI agent's actions affect its environment. Venture investors like Steve Jang at Kindred Ventures are backing this shift, viewing it as the next frontier beyond chatbot development.

What to watch

Fei-Fei Li has proposed a taxonomy dividing world models into three categories: 'renderers' (visual quality but limited robot training), 'simulators' (faithful physical representation for training), and 'planners' (predicting what an AI agent should do). According to Li, 'a robot that can plan is a robot that can work, and the entire industry is racing to be the one that gets there first.'