术语RLHF这些定义是英文。Reinforcement learning from human feedback: a training step that uses people’s preferences between outputs.