No PriorsNo Priors Ep. 117 | With Co-Director of Stanford's HAI & Founder of World Labs Dr. Fei-Fei Li
At a glance
WHAT IT’S REALLY ABOUT
Fei-Fei Li on spatial intelligence, world models, and fearless AI
- Dr. Fei-Fei Li discusses why she founded World Labs to build 3D world-model foundation models and argues that spatial intelligence is as fundamental to AI as language. She explains spatial intelligence as understanding, reasoning about, and generating plausible 3D environments that respect geometry and physics, enabling applications from robotics to creative tools and AR/VR. The conversation covers unsolved frontiers like emotional intelligence, the role of simulation and haptics in robotics, and the importance of diverse robot morphologies optimized for specific tasks. Li also reflects on her ImageNet and captioning work, advocates for fearless research and entrepreneurship outside mega-corporate labs, and articulates a vision of human-centered AI that augments people, especially in domains like healthcare.
IDEAS WORTH REMEMBERING
5 ideasSpatial intelligence and 3D world models are missing pillars of current AI.
Li argues that AI is incomplete without robust spatial intelligence—the capacity to understand, reason about, and generate 3D worlds that are geometrically and physically plausible, just as evolution endowed humans and animals with such capabilities.
Building 3D foundation models will unlock new classes of applications.
World Labs is focused on solving 3D generation as a foundation model problem, which could power applications in design, navigation, simulation, robotics, and immersive AR/VR/XR by providing realistic, editable 3D environments.
Data scarcity and productization are major challenges for 3D AI.
Unlike language models that benefit from abundant web text, 3D models require sophisticated data acquisition, synthesis, and engineering, and must overcome the friction of delivering 3D as an intuitive, everyday medium.
Simulation and haptics are undervalued components in training robots.
Li believes simulation and synthetic data are crucial for robotics, and that haptic sensing—how systems perceive touch and force—must be tightly integrated with vision and spatial perception to enable robust manipulation, not just navigation.
Robotic forms will diversify to optimize for task efficiency and energy.
She anticipates a wide variety of robot morphologies tailored to their environments and tasks (e.g., fish-like robots underwater, non-humanoid flying systems), driven by gradients of productivity and energy efficiency rather than a single humanoid standard.
WORDS WORTH SAVING
5 quotesWithout spatial intelligence, AI would be incomplete.
— Fei-Fei Li
We are the first company we know of that is solving this 3D generation foundation model problem.
— Fei-Fei Li
My hypothesis is that the requirements of different tasks are so vast that having very few forms is energy inefficient.
— Fei-Fei Li
Sometimes fearless is this very interesting position where you're somewhat delusional and crazy, but somewhat just rationally bold.
— Fei-Fei Li
I think AI is a tool to help people… I want to build a world that AI collaborates and superpowers people.
— Fei-Fei Li
High quality AI-generated summary created from speaker-labeled transcript.
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome