#839: Dr. Fei-Fei Li, The Godmother of AI — Asking Audacious Questions, Civilizational Technology, and Finding Your North Star

Hello, ladies and gentlemen and assorted adventurers—Tim Ferriss here, and welcome back to the show where I break down world-class performers so you can borrow what works. Today’s guest is Dr. Fei-Fei Li, Sequoia Professor of Computer Science at Stanford, cofounder of the Human-Centered AI Institute, CEO of World Labs, and author of The Worlds I See.

We realized we once shared a campus but never crossed paths, which is wild, so let’s rewind and start at the beginning—where did you grow up, and what shaped you?

I was born in Beijing, grew up mostly in Chengdu, and at fifteen moved with my mom to Parsippany, New Jersey, where we started over as new immigrants; I later studied physics at Princeton, then earned my PhD at Caltech in AI.

Your dad sounds delightfully unusual—less tiger parent, more playful explorer—what was he like?

He’s a curious, childlike lover of nature who took me to fields with water buffalo, chased bugs, and treated yard sales like treasure hunts, and he never cared about grades or trophies.

Why did your family leave China for the United States, given the language, financial, and cultural hurdles?

As a teen I didn’t fully grasp it, but now I see it as a brave leap for opportunity in a pre-internet world that felt like another planet, and I already had an unusual obsession with physics and fighter jets.

Where did your drive and technical focus come from, and who was Bob Zybala in your story?

My mom isn’t technical, but she’s disciplined, rebellious due to an education cut short, and she pushed focus without chasing awards; Bob was my high school calculus teacher who became family, tutored me through Calculus BC during his only lunch break, and showed me how much unsung public-school teachers give.

Let’s jump to ImageNet: what is it, why did it matter, and how did it begin?

Between 2007 and 2009 my student and I built ImageNet, a large-scale visual dataset that, combined with neural networks and GPUs, helped trigger the 2012 breakthrough in image recognition many see as the start of modern AI; my physics roots taught me to chase audacious questions, and mine became visual intelligence—how to make machines see and make sense of the world.

Why did you focus on data rather than just better models when others hadn’t?

Science is a lineage, not a lone-genius tale, and work like Irv Biederman’s on how children rapidly learn many objects helped crystallize the big-data hypothesis for vision.

What choices made ImageNet work, and how did you avoid bad incentives with crowd labeling?

We asked the right question—object categories instead of low-level pixels or overly broad scenes—then balanced quantity with quality, decided on image diversity, and, out of necessity, used Mechanical Turk with qualification quizzes and hidden gold-standard checks, filtering billions of images down to about fifteen million high-quality labeled examples.

Fast-forward to today: what’s missing in the AI conversation?

People are missing the people—AI is civilizational in its economic and cultural impact, but we must keep dignity, agency, and broad participation at the center rather than letting hype or fear drown them out.

Are we worrying too much, too little, or about the wrong things?

I’m a pragmatic optimist and I think the U.S. can lean into informed hope and inclusion, with technologists communicating better so communities across all fifty states see how they can benefit and shape what comes next.

Why build World Labs, and why now?

We’re building next-generation AI for spatial intelligence because seeing, reasoning, and doing in 3D unlocks creation, design, manufacturing, and robotics at a new level, and I believe advancing such tools is part of how humanity builds better civilizations.

Define spatial intelligence and what your product currently does.

Spatial intelligence is the loop between perceiving the world and acting within it, and our model, Marble, creates and reasons about interactive 3D worlds so creators and robots can simulate, explore, and make things.

Paint a picture with examples, and how would a nontechnical teacher actually use it?

A drama teacher could conjure a medieval town set, while VFX artists, interior designers, game makers, educators, therapists, and roboticists use it for scenes, studies, and simulation; you open a web page, type or upload a reference image, and a few minutes later you can walk through a navigable 3D scene, set cameras, and feed it into your workflow.

You mentioned psychiatric research—what might that look like?

Researchers can vary triggers and contexts on demand, creating controlled exposure environments—day or night, crowded or sparse—to study symptoms and test treatments far more flexibly and cheaply.

Who do you read or follow that deserves more attention?

I draw from cognitive science and robotics, including the late Anne Treisman and Irv Biederman, and I follow Liz Spelke, Alison Gopnik, and Rodney Brooks, among many others.

What near-term developments feel almost inevitable yet underappreciated?

Spatial intelligence will matter as much as language, AI will reshape education and credentials by accelerating self-directed learning, and labor impacts will be nuanced rather than apocalyptic or utopian.

What should young people focus on, and how do you hire now?

Learn how to learn and embrace AI collaboration tools, because adaptability and outcomes matter more than pedigree, and we look for engineers who grow with the fastest toolchains.

How should schools respond when AI can produce passable work?

If AI and student outputs are indistinguishable, the assignment is flawed, so show students the AI baseline and set a higher bar that rewards human thinking, synthesis, and originality.

If you had a global billboard, what would it say, and how do people find their North Star?

Ask yourself, what is your North Star, then use the journey of education—formal and informal—to learn who you are, define that star, and pursue it with intent.

One last thing—what’s the story behind your name, Fei-Fei?

Fei means fly, and my dad, biking to the hospital late for my birth, caught and released a bird, which inspired my name.

I had my own Chinese-name misadventures with tones back in the day, but we’ll save that for another time; thank you, Dr. Li—show notes are at tim.blog, and I appreciate you making the time.

Thank you—I loved the conversation, and thanks for having me.