Short Wave AI is great at predicting text. Can it guide robots?
64 snips
Feb 11, 2026 Geoff Brumfiel, an NPR science editor who visited robotics labs, reports on AI moving from text into physical robots. He describes lab demos of neural nets teaching robots simple tasks. He explores why robots still fail, how simulation and self-training could scale learning, and which practical, incremental uses are already working.
AI Snips
Chapters
Transcript
Episode notes
Stanford Lab Demo With Trail Mix
- Geoffrey Brumfiel watched a Stanford robot with two mechanical arms learn tasks using a model called OpenVLA.
- The robot learned by humans demonstrating actions repeatedly instead of being explicitly programmed.
Learning By Reinforcing Connections
- The robot used a teachable neural network that learns by reinforcing useful connections.
- Instead of coded instructions, the AI tunes billions of connections from repeated demonstrations.
Startup Robot That Folds Laundry
- Chelsea Finn's startup, Physical Intelligence, showed a mobile robot that removes laundry from dryers and folds it.
- The folding robot was trained by humans using powerful AI programs.