Dense Reward for Free in Reinforcement Learning from Human Feedback