RLHF Deciphered: A Critical Analysis of Reinforcement Learning from Human Feedback for LLMs