Skip to yearly menu bar Skip to main content


Self-Refine: Iterative Refinement with Self-Feedback

Aman Madaan · Niket Tandon · Prakhar Gupta · Skyler Hallinan · Luyu Gao · Sarah Wiegreffe · Uri Alon · Nouha Dziri · Shrimai Prabhumoye · Yiming Yang · Shashank Gupta · Bodhisattwa Prasad Majumder · Katherine Hermann · Sean Welleck · Sean Welleck · Amir Yazdanbakhsh · Peter Clark

Great Hall & Hall B1+B2 (level 1) #405
[ ] [ Project Page ]
[ Paper [ Slides [ OpenReview
Thu 14 Dec 3 p.m. PST — 5 p.m. PST

Abstract: Like humans, large language models (LLMs) do not always generate the best output on their first try. Motivated by how humans refine their written text, we introduce Self-Refine, an approach for improving initial outputs from LLMs through iterative feedback and refinement. The main idea is to generate an initial output using an LLMs; then, the same LLMs provides *feedback* for its output and uses it to *refine* itself, iteratively. Self-Refine does not require any supervised training data, additional training, or reinforcement learning, and instead uses a single LLM as the generator, refiner and the feedback provider. We evaluate Self-Refine across 7 diverse tasks, ranging from dialog response generation to mathematical reasoning, using state-of-the-art (GPT-3.5, ChatGPT, and GPT-4) LLMs. Across all evaluated tasks, outputs generated with Self-Refine are preferred by humans and automatic metrics over those generated with the same LLM using conventional one-step generation, improving by $\sim$20\% absolute on average in task performance. Our work demonstrates that even state-of-the-art LLMs like GPT-4 can be further improved at test-time using our simple, standalone approach.

Chat is not available.