Jun 4, 2025 · 22m · a16z

How Fei-Fei Li Is Rebuilding AI for the Real World

Fei-Fei Li · 13m spoken Martin Casado · 6m spoken Eric Vishria · 1m spoken
0:00 / 0:00
▶ Watch on YouTube →

gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions

In this episode of The a16z Podcast, AI pioneer Dr. Fei-Fei Li and General Partner Martin Casado discuss why artificial intelligence must evolve beyond language models to master 3D spatial intelligence. They detail the founding mission, scientific underpinnings, and real-world applications of World Labs in developing Large World Models.

How this conversation actually went

Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →

The host as informed peer 1.3 Guest teaching 3.0 Guest disagreement 0.5 The host pushing back 0.5
05100:0010:0020:000:39–5:10 · The host as informed peer 1/10 Introducing Fei-Fei Li: The Godmother of AI The host provides friendly setup questions introducing guest Fei-Fei Li and investor Martin Casado. The dynamic is purely collaborative and promotional as they share how World Labs was conceived.5:10–8:58 · The host as informed peer 2/10 Reflecting on a Decade of AI Progress The host asks why LLMs are not enough and why a new foundation model company is necessary. Fei-Fei educates on how language is a recent, lossy encoding of thought while spatial 3D intelligence is fundamental to living things.8:58–11:17 · The host as informed peer 1/10 Evolutionary Perspective on Language vs. Spatial Intelligence The guests outline the evolutionary timeline of vision versus language models. Fei-Fei gently corrects Martin's evolutionary estimate by pointing out spatial intelligence dates back 500 million years to trilobites.11:17–15:57 · The host as informed peer 1/10 Spatial Intelligence as the Engine of Human Breakthroughs The host prompts for concrete applications of world models. Fei-Fei and Martin walk through practical use cases in robotics, design, and generative multiverses.15:57–19:23 · The host as informed peer 2/10 Why 3D Representation Is Essential: The Tree Analogy and Monocular Vision Story The host questions why 2D representations cannot suffice. Fei-Fei illustrates the essential nature of 3D vision using an anecdote about losing her stereo vision after an eye injury and feeling terrified to drive.19:23–22:04 · The host as informed peer 1/10 The Scientific Foundations and Interdisciplinary Team of World Labs The host asks about the state of research in 3D AI relative to LLMs. Fei-Fei details the academic contributions of her co-founders across NeRFs, Gaussian splatting, and diffusion models.0:39–5:10 · Guest teaching 1/10 Introducing Fei-Fei Li: The Godmother of AI The host provides friendly setup questions introducing guest Fei-Fei Li and investor Martin Casado. The dynamic is purely collaborative and promotional as they share how World Labs was conceived.5:10–8:58 · Guest teaching 4/10 Reflecting on a Decade of AI Progress The host asks why LLMs are not enough and why a new foundation model company is necessary. Fei-Fei educates on how language is a recent, lossy encoding of thought while spatial 3D intelligence is fundamental to living things.8:58–11:17 · Guest teaching 4/10 Evolutionary Perspective on Language vs. Spatial Intelligence The guests outline the evolutionary timeline of vision versus language models. Fei-Fei gently corrects Martin's evolutionary estimate by pointing out spatial intelligence dates back 500 million years to trilobites.11:17–15:57 · Guest teaching 2/10 Spatial Intelligence as the Engine of Human Breakthroughs The host prompts for concrete applications of world models. Fei-Fei and Martin walk through practical use cases in robotics, design, and generative multiverses.15:57–19:23 · Guest teaching 4/10 Why 3D Representation Is Essential: The Tree Analogy and Monocular Vision Story The host questions why 2D representations cannot suffice. Fei-Fei illustrates the essential nature of 3D vision using an anecdote about losing her stereo vision after an eye injury and feeling terrified to drive.19:23–22:04 · Guest teaching 3/10 The Scientific Foundations and Interdisciplinary Team of World Labs The host asks about the state of research in 3D AI relative to LLMs. Fei-Fei details the academic contributions of her co-founders across NeRFs, Gaussian splatting, and diffusion models.0:39–5:10 · Guest disagreement 0/10 Introducing Fei-Fei Li: The Godmother of AI The host provides friendly setup questions introducing guest Fei-Fei Li and investor Martin Casado. The dynamic is purely collaborative and promotional as they share how World Labs was conceived.5:10–8:58 · Guest disagreement 1/10 Reflecting on a Decade of AI Progress The host asks why LLMs are not enough and why a new foundation model company is necessary. Fei-Fei educates on how language is a recent, lossy encoding of thought while spatial 3D intelligence is fundamental to living things.8:58–11:17 · Guest disagreement 1/10 Evolutionary Perspective on Language vs. Spatial Intelligence The guests outline the evolutionary timeline of vision versus language models. Fei-Fei gently corrects Martin's evolutionary estimate by pointing out spatial intelligence dates back 500 million years to trilobites.11:17–15:57 · Guest disagreement 0/10 Spatial Intelligence as the Engine of Human Breakthroughs The host prompts for concrete applications of world models. Fei-Fei and Martin walk through practical use cases in robotics, design, and generative multiverses.15:57–19:23 · Guest disagreement 1/10 Why 3D Representation Is Essential: The Tree Analogy and Monocular Vision Story The host questions why 2D representations cannot suffice. Fei-Fei illustrates the essential nature of 3D vision using an anecdote about losing her stereo vision after an eye injury and feeling terrified to drive.19:23–22:04 · Guest disagreement 0/10 The Scientific Foundations and Interdisciplinary Team of World Labs The host asks about the state of research in 3D AI relative to LLMs. Fei-Fei details the academic contributions of her co-founders across NeRFs, Gaussian splatting, and diffusion models.0:39–5:10 · The host pushing back 0/10 Introducing Fei-Fei Li: The Godmother of AI The host provides friendly setup questions introducing guest Fei-Fei Li and investor Martin Casado. The dynamic is purely collaborative and promotional as they share how World Labs was conceived.5:10–8:58 · The host pushing back 2/10 Reflecting on a Decade of AI Progress The host asks why LLMs are not enough and why a new foundation model company is necessary. Fei-Fei educates on how language is a recent, lossy encoding of thought while spatial 3D intelligence is fundamental to living things.8:58–11:17 · The host pushing back 0/10 Evolutionary Perspective on Language vs. Spatial Intelligence The guests outline the evolutionary timeline of vision versus language models. Fei-Fei gently corrects Martin's evolutionary estimate by pointing out spatial intelligence dates back 500 million years to trilobites.11:17–15:57 · The host pushing back 0/10 Spatial Intelligence as the Engine of Human Breakthroughs The host prompts for concrete applications of world models. Fei-Fei and Martin walk through practical use cases in robotics, design, and generative multiverses.15:57–19:23 · The host pushing back 1/10 Why 3D Representation Is Essential: The Tree Analogy and Monocular Vision Story The host questions why 2D representations cannot suffice. Fei-Fei illustrates the essential nature of 3D vision using an anecdote about losing her stereo vision after an eye injury and feeling terrified to drive.19:23–22:04 · The host pushing back 0/10 The Scientific Foundations and Interdisciplinary Team of World Labs The host asks about the state of research in 3D AI relative to LLMs. Fei-Fei details the academic contributions of her co-founders across NeRFs, Gaussian splatting, and diffusion models.

speaking balance: gold is the host, purple is the guest (3 minute bins)

0:00 · the host 0% · guest 100%0:00 · the host 0% · guest 100%3:00 · the host 0% · guest 100%3:00 · the host 0% · guest 100%6:00 · the host 0% · guest 100%6:00 · the host 0% · guest 100%9:00 · the host 0% · guest 100%9:00 · the host 0% · guest 100%12:00 · the host 0% · guest 100%12:00 · the host 0% · guest 100%15:00 · the host 0% · guest 100%15:00 · the host 0% · guest 100%18:00 · the host 0% · guest 100%18:00 · the host 0% · guest 100%21:00 · the host 0% · guest 100%21:00 · the host 0% · guest 100%
Sharpest disagreement ▶ 10:14 Correcting the evolutionary timeline

In a highly amicable interview, the closest moment to friction occurs when Fei-Fei interrupts Martin to correct his evolutionary timeline, clarifying that spatial vision dates back 500 million years to trilobites rather than 4 million.

Hardest push from the host ▶ 6:01 Challenging the need beyond LLMs

The host directly asks why LLMs are not enough and questions the rationale for building another foundation model company when language models are dominating.

Biggest teaching moment ▶ 6:05 Reframing language vs spatial intelligence

Fei-Fei reframes the entire AI paradigm by explaining that language is merely a recent, lossy human construct, whereas 3D spatial perception is the true foundation of embodied intelligence.

The host holds their own ▶ 16:51 Probing 2D limitations

The host tests the technical premise of world models by asking why existing 2D video outputs cannot simply be used instead of dedicated 3D representations.

the scores for every segment, with the reasoning behind each
ChapterTopicThe host as informed peerGuest teachingGuest disagreementThe host pushing backWhy
Introducing Fei-Fei Li: The Godmother of AI 1100 The host provides friendly setup questions introducing guest Fei-Fei Li and investor Martin Casado. The dynamic is purely collaborative and promotional as they share how World Labs was conceived.
Reflecting on a Decade of AI Progress 2412 The host asks why LLMs are not enough and why a new foundation model company is necessary. Fei-Fei educates on how language is a recent, lossy encoding of thought while spatial 3D intelligence is fundamental to living things.
Evolutionary Perspective on Language vs. Spatial Intelligence 1410 The guests outline the evolutionary timeline of vision versus language models. Fei-Fei gently corrects Martin's evolutionary estimate by pointing out spatial intelligence dates back 500 million years to trilobites.
Spatial Intelligence as the Engine of Human Breakthroughs 1200 The host prompts for concrete applications of world models. Fei-Fei and Martin walk through practical use cases in robotics, design, and generative multiverses.
Why 3D Representation Is Essential: The Tree Analogy and Monocular Vision Story 2411 The host questions why 2D representations cannot suffice. Fei-Fei illustrates the essential nature of 3D vision using an anecdote about losing her stereo vision after an eye injury and feeling terrified to drive.
The Scientific Foundations and Interdisciplinary Team of World Labs 1300 The host asks about the state of research in 3D AI relative to LLMs. Fei-Fei details the academic contributions of her co-founders across NeRFs, Gaussian splatting, and diffusion models.

Statements from this episode (13)

Insight
Li: Spatial intelligence is a critical component of intelligence
“Space, the three D space, the space out there, the space in your mind's eye, the spatial intelligence is a part of a critical part of intelligence.”
Fei-Fei Li Jun 4, 2025 ▶ 0:00
Prediction Not checkable as stated
Li: Spatial AI will enable humanity to live in a multiverse
“Suddenly we can actually create infinite universes. Some are for robots, some are for creativity, some are for socialization, some are for travel, some are for storytelling. It suddenly will enable us to live in a multiverse.”
Fei-Fei Li Jun 4, 2025 ▶ 0:11
Assertion Not checkable as stated
Casado: Fei-Fei Li singularly brought data scale to AI
“But Fei-Fei really singularly brought in data to the equation, which now we're recognizing is actually probably the bigger problem, the more interesting one,”
Martin Casado Jun 4, 2025 ▶ 1:07
Disclosure
Li: Most investors and technologists initially failed to grasp world models
“I was at that point already talking to various people, not just computer scientists, technologists, but also investors and potentially business partners, and to be honest, most people didn't get it.”
Fei-Fei Li Jun 4, 2025 ▶ 4:11
Opinion
Li: Data-driven AI models exhibit emergent thinking behaviors
“Data hungry models, the data driven AI can come this far and genuinely have incredible emergent behaviors of thinking machine”
Fei-Fei Li Jun 4, 2025 ▶ 5:34
Insight
Fei-Fei Li: Language is a lossy representation of the physical world
“Language is a lossy way to capture the world.”
Fei-Fei Li Jun 4, 2025 ▶ 6:46
Assertion Not checkable as stated
Fei-Fei Li: Spatial AI now requires industry-scale compute and talent
“The time has come that concentrated industry Grade effort, focus effort in terms of compute data talent is, is really the answer to bringing this to life.”
Fei-Fei Li Jun 4, 2025 ▶ 7:33
Assertion Supported
Casado: Tech industry has invested $100 billion in autonomous vehicles
“Even to look at autonomous vehicles, I mean, as an industry, we've invested, like, what, a hundred billion dollars in it.”
Martin Casado Jun 4, 2025 ▶ 9:17
Assertion Not checkable as stated
Casado: LLMs are unit economic positive and quickly solved language problems
“And then out of nowhere comes these LLMs, and they are unit economic positive, they solve all of these language problems, like, basically immediately.”
Martin Casado Jun 4, 2025 ▶ 9:42
Insight
Fei-Fei Li: All embodied robotics require 3D spatial intelligence
“Robotics to me is any body machines. It's not just humanoids or cars. There's so much in between, but all of them have to somehow figured out, ah, the three D space it lives in, have to be trained, ah, to understand the three D space and have to do things, som…”
Fei-Fei Li Jun 4, 2025 ▶ 13:04
Insight
Casado: World models are horizontal tech converting 2D inputs into 3D action
“It's a super, super horizontal piece that takes basically a computer with a single view in the world, or maybe multiple views in the world, and creates a full three D representation that that computer then can act on. And so you can see that that's a very, lik…”
Martin Casado Jun 4, 2025 ▶ 15:35
Insight
Fei-Fei Li: AI World Modeling Is Fundamentally a 3D Problem
“Physics happens in three D and interaction happens in three D. Navigating behind the back of the table needs to happen in three D. Composing the world, whether physically, digitally needs to happen in three D. So fundamentally the problem is a three D problem.”
Fei-Fei Li Jun 4, 2025 ▶ 16:55
Insight
Casado: Solving 3D world models requires combining AI and graphics expertise
“To solve this problem, you need experts both in AI, and that's like the data and the models, like the actual model architecture and graphics, which is like, how do you actually represent these things in memory, in a computer, and then on the screen?”
Martin Casado Jun 4, 2025 ▶ 21:38
Made with StarZero

Turn any episode into a week of clips.

This entire site, over 1,000 episodes transcribed, diarized, checked and made playable, runs on the StarZero media pipeline. Drop in your own episode and the podcast clipper finds the moments worth sharing, cuts them, captions them, and reframes them for every feed.