The speaker's proposed training recipe combines a large source of human egocentric data, which provides broad physical-manipulation priors, with a small amount of robot data for task mastery.

“So the kind of recipe we're thinking about is like, how do we take the largest source of data we have today,”

“like egocentric data,”

“to give the model a strong prior over everything physical manipulation can cover,”

“and then combine that with a small amount of robotics data to make the robot master whatever task we're interested in. So data quality...”