Top developers are shifting from chatbots to physical AI. Here’s why
TL;DR
More AI builders are looking beyond chatbots toward world models: systems meant to understand space, objects, time, causality and the physical consequences of actions. Overworld founder Louis Castricato is used as the emblem of that shift from LLM-only work to models that can plan inside robotics, games, simulations and real environments. The piece points to support from Fei-Fei Li and Yann LeCun, but the field is still early, expensive and wrapped in plenty of investor-friendly framing.
Nauti's Take
World models are the better narrative for the AI industry right now: bigger than chat, closer to robotics, games, industry and autonomous action. But the gap between demo and daily use is brutal here.
Anyone treating physical AI as already solved is selling the future as a product. Still, the direction matters: the strongest AI workflows will likely be hybrid systems where LLMs understand language and intent, while world models simulate the environment, risks and next moves.
Briefingshow
The interesting point is not that chatbots go away. The next layer starts where AI stops continuing language and starts predicting state: what happens when a robot grabs, a car swerves or a character acts inside a game world. That is where AI either stays a screen tool or moves into real operating environments.
Video
Sources