English: This is a high-level overview of reinforcement learning from human feedback, including training an initial supervised model, collecting human feedback, training a reward model, and using it to align the initial model.