If a robot performs the same task every day, why invest so much in pre-training?
@SkildAI’s S1 suggests an answer: pre-training can buy deployment speed, not just better performance. In one experiment, a robot began attempting a new task only five minutes after watching a human
The robot data market likes to measure supply in hours but the most valuable company may not be the one that collects the most data.
What's the key & How different companies like @scale_AI@encord_team@micro1_ai@MeckaAI and @hbr_pbc are approaching different parts of this