Top developers are shifting from chatbots to physical AI. Here’s why
TL;DR
More top developers are looking beyond chatbots: world models aim to teach AI space, time, objects, forces, and the consequences of its own actions. Louis Castricato left his LLM PhD at Brown University and founded Overworld, a startup building interactive world models for game environments instead of another text assistant. Fei-Fei Li, Yann LeCun, and robotics researchers see this as a base layer for physical AI: systems that can act in real or simulated environments, not just predict words or pixels.
Nauti's Take
The chatbot was the interface for thinking, not doing. If you're still just polishing prompts, you're missing the harder layer: models that understand space, risk, and consequences.
For builders, this means less demo magic and more sensors, simulation, and reliable feedback loops.
Briefingshow
The shift points to AI’s next hard boundary: language is enough for office workflows, but not for robots, games, or agents that must act reliably in messy environments. World models could become the bridge between LLMs and physical automation. At the same time, the term is already overloaded, covering everything from pretty renderers to real planning systems.
Video
Sources