On a Connection Between Imitation Learning and RLHF