Fei-Fei Li, Justin Johnson, and Ben Mildenhall on World Models and Spatial Intelligence
In this episode of The a16z Show, World Labs co-founders Fei-Fei Li, Justin Johnson, and Ben Mildenhall talk with Martin Casado about Atlas, their new world model aimed at advancing spatial intelligence. Central to Atlas is new view prediction, an approach that predicts what an environment looks like from different positions in space and time given existing views or images. This methodology integrates generation and 3D reconstruction into a unified model, exploring whether view prediction can serve as a fundamental primitive for understanding physical reality. The founders examine the technical bets underpinning Atlas, detailing current capabilities, limitations, and the necessity of dynamics, editability, and simulation in future world models. Additionally, they discuss practical applications in architecture, creative industries, and robotics, where Fei-Fei Li notes that access to real-world training data remains a critical bottleneck.
Original sources
Podcast The a16z Show
AI Radar wrote this summary from the publisher’s notes with AI, not from a transcript. How it works.