Fei-Fei Li: World Models and the Multiverse
The a16z Show
2025/06/04
Fei-Fei Li: World Models and the Multiverse
Fei-Fei Li: World Models and the Multiverse

The a16z Show
2025/06/04
This podcast delves into the future of artificial intelligence by exploring spatial intelligence and its potential to revolutionize AI beyond language models. Fei-Fei Li, a pioneer in AI, discusses her vision for bridging the gap between language and spatial understanding with Martin Casado, an early supporter of her work.
Fei-Fei Li highlights the importance of finding the right investor for her venture, emphasizing intellectual alignment over mere financial backing. She advocates for developing world models that go beyond language to represent the physical world accurately. By focusing on spatial intelligence, AI can enhance creativity, robotics, and even create virtual universes. The conversation explores why advancements in language models have outpaced robotics and how incorporating 3D concepts like DNA structures could transform various fields. Li shares personal insights about navigating the world with stereo vision, underscoring the significance of 3D perception for interaction and physics. New technologies such as neural radiance fields and Gaussian splat representation are also discussed, showcasing emerging research areas in 3D computer vision.
00:00
00:00
Fei-Fei Li is called the 'godmother of AI' by Martin Casado.
05:15
05:15
LLMs are insufficient as language is a lossy way to represent the 3D physical world.
10:00
10:00
Navigating the real world relies on spatial reconstruction
11:02
11:02
Spatial intelligence extends AI capabilities beyond LLMs with applications in creativity and robotics.
14:14
14:14
Combination of generation and reconstruction technology enables infinite multiverse.