
Bitter Lesson Limitations for Robotics

The Godmother of AI on jobs, robots & why world models are next | Dr. Fei-Fei Li
Topic 6 of 9
Published Nov 16, 2025 · Clip 41:16–48:06 (6:49)
The bitter lesson (simple models + massive data outperform complex ones) drove language AI but faces robotics hurdles: data scarcity (web videos lack 3D actions), misalignment (videos vs. needed actions), and physical embodiment like self-driving cars' 20-year maturation. Solutions include teleoperation/synthetic data and world models for simulation. Robotics is early-stage, needing brains, bodies, and scenarios; closer to cars (avoid touching) than LLMs, requiring multiple bitter lessons.