Large Language Models have Intrinsic Self-Correction Ability

TL;DR

This paper investigates the intrinsic self-correction ability of large language models, showing that with proper settings like zero temperature and fair prompts, LLMs can effectively revise their answers without external knowledge.

Contribution

The paper provides a theoretical and empirical analysis demonstrating intrinsic self-correction in LLMs and highlights key factors like zero temperature and fair prompts for success.

Findings

Intrinsic self-correction exists across multiple LLMs.

Zero temperature and fair prompts are critical for effective self-correction.

Insights into fundamental theories of LLM self-correction behavior.

Abstract

Large language models (LLMs) have attracted significant attention for their exceptional abilities in various natural language processing tasks, but they suffer from hallucinations that will cause performance degradation. One promising solution to improve the LLMs' performance is to ask LLMs to revise their answer after generation, a technique known as self-correction. Among the two types of self-correction, intrinsic self-correction is considered a promising direction because it does not utilize external knowledge. However, recent works doubt the validity of LLM's ability to conduct intrinsic self-correction. In this paper, we present a novel perspective on the intrinsic self-correction capabilities of LLMs through theoretical analyses and empirical experiments. In addition, we identify two critical factors for successful self-correction: zero temperature and fair prompts. Leveraging these factors, we demonstrate that intrinsic self-correction ability is exhibited across multiple existing LLMs. Our findings offer insights into the fundamental theories underlying the self-correction behavior of LLMs and remark on the importance of unbiased prompts and zero temperature settings in harnessing their full potential.