TechCrunch: Physical AI Training Data Is Moving Past Video Toward Multi-Camera Rigs and Brain-Wave Capture
TechCrunch·low signal
TechCrunch reported July 26 that frontier physical-AI models have exhausted what scraped YouTube video can teach them and now require purpose-collected data — multiple synchronized camera angles, dense annotation, and, increasingly, EEG brain-wave readings from human demonstrators. The argument is that intent and attention aren't recoverable from pixels alone, so the labs are instrumenting the humans rather than just the scene. Treat as a directional signal rather than a confirmed capability result: it's a single-outlet trend piece, not a paper or product announcement.