World Labs' Atlas: new-view prediction rebuilds 3D spaces from 3 photos, 50-100x cheaper
drfeifei · x · 2026-09-06
In an a16z conversation, World Labs co-founders Fei-Fei Li, Justin Johnson and Ben Mildenhall detail Atlas, a world model for spatial intelligence.
- Core thesis: LLMs rest on next-token prediction and video models on next-frame prediction; Atlas is built on generative new-view prediction, which Johnson argues is also AI-complete — solving it in full generality would solve any intelligence problem.
- Technical milestone: Atlas is the first model to unify pixel generation and pixel reconstruction, two problems computer vision has kept separate for half a century.
- Practical payoff: digitally capturing a 3D space previously needed 100-300 photos per room; Atlas works from just three, a 50-100x cost reduction.
- The discussion also touches on implications for robotics and why the team emphasizes spatial reasoning.
More from Companies & People
- a16z's Casado says AI can't do a VC's job — but it can do everything around it — alvelda · 2026-10-21
- Ex-Meta Integrity engineer praises Zuck's shift from fact-checkers to Community Notes — HarperSCarroll · 2026-10-21
- Nyne, under 15 people with tens of millions in monthly revenue, hires SPLs and RL engineers — soleio · 2026-10-21
- UNESCO and ICOM Release 2026 Report on AI Use in Museums Across 90 Countries — LuizaJarovsky · 2026-10-21
- AI app-layer acquisitions are now mostly about founders and customers, not tech — signulll · 2026-10-21
- Vanta CEO on the 'boring and complex' playbook, eating AI costs, and 10X comp — soleio · 2026-10-21