Reinforcement learning, I'd assume.

Presumably RLHF (Reinforcement Learning from Human Feedback).