The PodKnowledge|← All Episodes
Based on Lenny's Podcast data
Lenny's Knowledge Sketch

The Godmother of AI on
What Comes Next

Dr. Fei-Fei Li
Co-director Stanford HAI; Godmother of AI; founder World Labs
NOV 16 2025
The Vision

World Models Are
the Next Frontier

AI CAPABILITY
"So much of our intelligence is built upon visual, perceptual, spatial understanding, not just language per se. I think they're complementary."
  • World Labs' mission: build AI with spatial intelligence, not just language intelligence
  • ImageNet moment (2012) sparked the deep learning revolution, we're at a new inflection
  • Spatial AI enables robots, autonomous systems, and physical world automation
  • The gap: current AI understands text and images but not 3D space and physical causality
Framework

From ImageNet to World Models

PERCEPTIONLANGUAGESPATIAL WORLD
2012
ImageNet breakthrough
2017
Companies start calling themselves AI
  • ImageNet: AI learns to see (perception)
  • GPT: AI learns to speak and reason (language)
  • World models: AI learns to understand and predict physical reality
  • Each leap required new architecture, new data, new compute scale
Fei-Fei's thesisWorld modeling and spatial intelligence is a key missing piece of embodied AI, and complementary to language models.
The Jobs and Society Question

What AI Means for Humans

  • Fei-Fei's optimism: AI augments human capability, in medicine, education, creativity
  • Her concern: Every technology is a double-edged sword. If we're not doing the right thing as a society, we can screw this up too.
  • On jobs: Humans are embodied agents too. We can be augmented by AI, just as language AI already augments us at work today.
The HAI mission

Human-Centered AI: technology designed around human flourishing, not just capability maximization.

The responsibility principle

No matter which part of AI you build, deploy, or use, act like a responsible individual and care a lot about this.

Playbook

Think Like an AI Pioneer

  • Study the physical world problem space, that's where the next generation of AI impact is
  • Ask: does this AI system increase or decrease human agency? It should always increase it.
  • Be intellectually fearless. If you want to make a difference, you have to accept diving into something no one has done
  • Anchor AI in human benevolence and human centeredness, not just capability
The ImageNet lessonBig data, neural network, and GPU: the golden recipe for modern AI. Still the three ingredients at its core.
Contrarian

AI Pessimism Myths Fei-Fei Addresses

✗AI will make humans obsoleteINSTEAD →✓ Humans are embodied agents who can be augmented by AI, just as language AI already augments us at work today.
✗AGI is a real scientific milestone that's almost hereINSTEAD →✓ AGI is more a marketing term than a scientific term. AI is still one of the youngest disciplines: we're scratching the surface.
✗Robots will scale as fast as language models didINSTEAD →✓ Robots are closer to self-driving cars than to language models. They need brains, physical bodies, and application scenarios.
✗More data, more GPUs, bigger models will get us thereINSTEAD →✓ Scaling still has a lot to do, but we absolutely need to innovate more. AI is nowhere near done.
Based on Dr. Fei-Fei Li's episode on Lenny's Podcast. All ideas on this page are from the episode.Watch on YouTubeFollow @drfeifei on X
GO DEEPER IN THE EPISODE
05:31 The evolution of AI09:37 The birth of ImageNet17:25 The rise of deep learning23:53 The future of AI and AGI29:51 Introduction to world models40:45 The bitter lesson in AI and robotics48:02 Introducing Marble, a revolutionary product51:00 Applications and use cases of Marble01:01:01 The founder’s journey and insights01:10:05 Human-centered AI at Stanford01:14:24 The role of AI in various professions01:18:16 Conclusion and final thoughts
Every sketch here began as a transcript.

Send us yours and we’ll draw it. $1 a transcript.

Sketch mine →
𝕏︎ X / Twitterin LinkedIn📸 Instagram🔗 Copy link
0:00