Chain-of-thought supports long-horizon tasks, and about six months before the interview the team found that its models could be improved simply by supervising them with high-level semantic labels on existing experience rather than adding more teleoperated low-level actions, shifting the bottleneck from low-level motor skill to mid-level scene interpretation and step selection, so a person can coach the robot by talking to it.
causalpending
Speaker
Sergey LevineEvidence Quote
“the bottleneck had actually shifted from the lowest level... to this like middle level”
Created: 11/11/2026, 5:04:53 AM
Your notes
Loading notes...