Self-Refine: Iterative Refinement with Self-Feedback

Self-Refine is an approach for improving initial outputs from large language models through iterative feedback and refinement, motivated by how humans revise their written text. A single LLM serves as generator, feedback provider, and refiner: it produces an initial output, gives feedback on that output, and uses the feedback to refine itself, repeating the loop. The method requires no supervised training data, no additional training, and no reinforcement learning.

Evaluated across 7 diverse tasks ranging from dialog response generation to mathematical reasoning, using state-of-the-art LLMs including GPT-3.5, ChatGPT, and GPT-4, Self-Refine outputs were preferred by both humans and automatic metrics over conventional one-step generation, improving task performance by about 20% absolute on average. The result mattered because it demonstrated that even top-tier models like GPT-4 can be further improved at test time using a simple, standalone technique without any retraining.