Teaching Robots To Think

Sergey explains how their vision-language-action models reason before acting—using chain-of-thought to unlock web-scale common sense, then reinforcement learning to get faster and more robust in the real world.