Fei-Fei Li

Who

Fei-Fei Li moved from China to the United States as a teenager, studied physics at Princeton and earned her PhD at Caltech. She led the creation of ImageNet, a large labeled image dataset whose annual challenge helped spark the deep learning era. She is a professor at Stanford and co-directs its Institute for Human-Centered AI. In 2024 she co-founded World Labs to work on spatial intelligence, and her memoir is The Worlds I See.

Operating rules

Fix the data before the algorithm. When most researchers were refining models, she bet that the field lacked enough good data. ImageNet took years of patient, unglamorous work. It became the ground that much of modern computer vision was built on.

Formulate the problem first. She often stresses that choosing the right question matters more than polishing an answer to the wrong one. A clear North Star also tells you what to say no to.

Ground intelligence in the physical world. Her current work starts from the view that language alone is not understanding. To be truly useful, AI needs to perceive and reason about three-dimensional space.

Keep people at the center. She argues that AI should be built to augment human abilities and that the field needs more diverse voices. She co-founded AI4ALL to bring more young people into it.

What I borrowed

Eleven years of BIM research taught me that a building model is only as good as its data. ImageNet showed the same lesson at a global scale. Her work on spatial intelligence asks, at a much larger scale, a question my own work keeps returning to: how to describe a physical space so that a machine can act on it.

What I won’t copy

The research time horizon. Building a foundation for years before it pays off suits a lab. In operations I need to show value each quarter, so I build foundations in smaller pieces.