
AI + a16z Fei-Fei Li: World Models and the Multiverse
43 snips
Dec 23, 2025 Fei-Fei Li, cofounder and CEO of World Labs, is a visionary in computer vision and AI. In this discussion, she reveals the crucial role of spatial intelligence in AI, explaining why world models can transform robotics and creativity. Reflecting on her journey, she describes the unique partnership with early investor Martin Casado. They delve into the limitations of language models versus the necessity of 3D understanding, and explore exciting applications for world models, paving the way for multiverse experiences in both real and virtual realms.
AI Snips
Chapters
Transcript
Episode notes
Language Is Lossy For The Physical World
- Language models excel but are lossy for representing physical reality.
- Spatial intelligence and 3D world models capture information language cannot and are core to broader intelligence.
Concentrate Resources To Productize World Models
- Concentrate industry-grade compute, data, and talent to build world models.
- Move beyond academic papers when a problem requires focused productization and scale.
Reconstruction Beats Description For Action
- Spatial reconstruction provides precise, actionable world understanding that language descriptions cannot.
- Humans reconstruct 3D from perception; AI needs explicit 3D representations to act in the world.

