Dr. Fei-Fei Li, The Godmother of AI — Asking Audacious Questions & Finding Your North Star

Fei-Fei Li traces her path from a middle-class childhood in Chengdu, China to immigrating to New Jersey at 15, running a family dry cleaning shop, and studying physics at Princeton before pursuing AI at Caltech. She explains how ImageNet emerged from a hypothesis that big data, not just better algorithms, was the missing ingredient for visual intelligence, and how it converged with neural networks and GPUs in 2012 to spark modern AI. She stresses that scientific progress is a non-linear lineage of many contributors rather than the work of lone geniuses. Turning to the present, she argues AI is a civilizational technology and that people, dignity, and agency are being underappreciated in the hype. She describes World Labs and its 'Marble' model for spatial intelligence, plus how AI should reshape education by raising the bar for human learners rather than policing tool use.

• ImageNet, combined with neural network algorithms and GPUs, converged in the 2012 'ImageNet classification deep convolutional neural network' paper that many call the birth of modern AI.
  00:19:17
• Amazon Mechanical Turk was adopted 'out of desperation' because labeling required tens of millions of human-filtered images that Princeton undergrads were too expensive and slow to handle.
  00:33:25
• Li argues the biggest thing people are missing in AI is the importance of people, dignity, and agency, framing AI as a civilizational technology.
  00:39:35
• World Labs is building next-generation AI focused on spatial intelligence, which Li calls a 'lynchpin technology' as fundamental as language intelligence.
  00:45:54
• Li says spatial intelligence and world modeling of 3D pixels are underappreciated while everyone focuses on large language models.
  00:58:01
• At World Labs in 2025, Li would not hire any software engineer who does not embrace AI-collaborative software tools.
  01:02:10
• Li contends that if a school evaluation produces the same result from AI and from a student, the evaluation structure itself is broken.
  01:03:44

• Li was born in Beijing and spent most of her childhood in Chengdu, famous for panda bears, before immigrating to Parsippany, New Jersey at age 15.
  00:03:08
• Her high school math teacher Bob Sabella sacrificed his only lunch hour to teach her one-on-one Calculus BC, and his family became her American family.
  00:14:07
• Li's family operated a dry cleaning shop in New Jersey, which she ran for seven years.
  00:16:41
• ImageNet was built between 2007 and 2009 as Li transitioned from assistant professor at Princeton to Stanford.
  00:17:13
• The first Mechanical Turk task Li tried herself was transcribing wine bottle labels.
  00:35:30
• ImageNet distilled billions of images down to 15 million high-quality labeled images.
  00:36:00
• Li cites an unverified figure that 50% of US GDP growth last year was attributed to AI, with overall growth at 4% versus 2% without AI.
  00:40:07
• World Labs' model is called 'Marble' and can generate 3D worlds, such as a medieval French town, from a text prompt or uploaded image.
  00:50:37
• A psychology researcher used World Labs' immersive worlds to study triggers and treatment for conditions like obsessive-compulsive disorder.
  00:54:51
• Li's name 'Fei' means 'flying,' inspired by her father catching a bird while bicycling late to her mother's hospital during labor.
  01:07:52

Books, products and media the guest or host genuinely endorsed here — with the buy link.

Affiliate link — we may earn a commission at no extra cost to you.