Refining the Responses of LLMs by Themselves

TL;DR

This paper presents a prompt engineering method enabling large language models to self-refine their responses iteratively, improving answer quality without auxiliary models, and achieving results comparable to or better than GPT-4.

Contribution

The paper introduces a self-evaluation based iterative refinement framework for LLMs that enhances response quality without external tools.

Findings

Self-refinement improves GPT-3.5 responses significantly.

The method achieves performance comparable to GPT-4.

Iterative process reduces manual intervention.

Abstract

In this paper, we propose a simple yet efficient approach based on prompt engineering that leverages the large language model itself to optimize its answers without relying on auxiliary models. We introduce an iterative self-evaluating optimization mechanism, with the potential for improved output quality as iterations progress, removing the need for manual intervention. The experiment's findings indicate that utilizing our response refinement framework on the GPT-3.5 model yields results that are on par with, or even surpass, those generated by the cutting-edge GPT-4 model. Detailed implementation strategies and illustrative examples are provided to demonstrate the superiority of our proposed solution.