[NeurIPS 2024] Self-Refine: Iterative Refinement with Self-Feedback

Contents

• TL;DR
• References

TL;DR

• 通过让 LLM 生成 feedback 不断 refine 自身的回答，可以提升回答效果，但也会带来不可忽视的推理开销

References

• Madaan, Aman, et al. "Self-refine: Iterative refinement with self-feedback." Advances in Neural Information Processing Systems 36 (2024).
• code : https://selfrefine.info/