Large Language Models Can Self-Correct with Key Condition Verification

TL;DR

This paper introduces ProCo, a simple verification framework that enhances large language models' reasoning accuracy by iteratively verifying and correcting responses through key condition verification, significantly outperforming previous self-correction methods.

Contribution

ProCo leverages minimal prompting to verify responses by checking key conditions, enabling effective self-correction in LLMs across multiple reasoning tasks.

Findings

+6.8% exact match on open-domain QA datasets

+14.1% accuracy on arithmetic reasoning datasets

+9.6% accuracy on commonsense reasoning

Abstract

Intrinsic self-correct was a method that instructed large language models (LLMs) to verify and correct their responses without external feedback. Unfortunately, the study concluded that the LLMs could not self-correct reasoning yet. We find that a simple yet effective verification method can unleash inherent capabilities of the LLMs. That is to mask a key condition in the question, add the current response to construct a verification question, and predict the condition to verify the response. The condition can be an entity in an open-domain question or a numeric value in a math question, which requires minimal effort (via prompting) to identify. We propose an iterative verify-then-correct framework to progressively identify and correct (probably) false responses, named ProCo. We conduct experiments on three reasoning tasks. On average, ProCo, with GPT-3.5-Turbo as the backend LLM, yields $+6.8$ exact match on four open-domain question answering datasets, $+14.1$ accuracy on three arithmetic reasoning datasets, and $+9.6$ accuracy on a commonsense reasoning dataset, compared to Self-Correct. Our implementation is made publicly available at https://wzy6642.github.io/proco.github.io/.