The Godmother of AI on jobs, robots & why world models are next

Dr. Fei-Fei Li, Stanford professor and co-creator of ImageNet, is widely called the godmother of AI. The conversation traces the history of modern AI from the AI winter through ImageNet and the 2012 deep learning breakthrough, then covers her views on AGI, the limits of current models, the bitter lesson for robotics, and her startup World Labs and its world model product Marble.

01Key learnings

• Large labeled datasets were the missing ingredient that unlocked modern AI; big data, neural networks, and GPUs together formed the recipe that still underlies today's systems.
• Pursue a north-star problem and commit to it for years, as ImageNet's object recognition focus did for her lab.
• Be willing to take intellectual and career risks, such as leaving tenure track or joining new ventures, when the mission and people are compelling.
• Choose opportunities based on passion and mission rather than over-optimizing every detail, which can paralyze people in a fast-moving field.
• Spatial intelligence and world models are a key missing piece for embodied AI and robotics, since language data lacks the 3D action data robots need.
• Robotics will likely take far longer than language models, as self-driving cars did, because physical systems need bodies, data, supply chains, and real application scenarios.
• Prototype and release early to discover real use cases; Marble's users surfaced unexpected applications like VFX production, simulation, and psychology research.
• Treat AI as a tool that should preserve human dignity and agency, and stay engaged in how it is governed and applied in your community.

02Frameworks mentioned

Summary and takeaways written by PM Atlas; quotes are short excerpts. © the original author.