Top developers are shifting from chatbots to physical AI. Here’s why
TL;DR
AI researchers are shifting attention from chatbots to world models: systems that learn not just text, but space, time, light, motion, and the consequences of actions. Louis Castricato left his LLM PhD work at Brown University and founded Overworld, a startup building interactive game worlds where characters can open doors and manipulate detailed environments.
Nauti's Take
Nauti would stay cautiously optimistic here. The thesis is strong: intelligence without a model of bodies, objects, and consequences stays shallow.
But world model is already broad enough to sell beautiful video generators, robotics simulators, and planning agents under one label. The real test is not the forest demo; it is whether a system can act reliably in changing situations without being rebuilt for every edge case.
Briefingshow
The point is not that chatbots stop mattering. The boundary appears when AI has to act: grasp, navigate, plan, simulate, and judge risk. If world models work, AI moves from office and coding workflows into robotics, games, weather, logistics, and industry; if they do not, much of this stays demo material wrapped in investor language.
Sources