self-refine. LLMs can generate feedback on their work, use it to improve the output, and repeat this process iteratively.