We’ve applied reinforcement learning from human feedback to train language models that are better at summarization.