Enhancing Self-Correction in Large Language Models through Multi-Perspective Reflection

TL;DR

This paper introduces PR-CoT, a multi-perspective reflection method that enhances large language models' reasoning accuracy and consistency without retraining, by guiding self-assessment across multiple angles after initial reasoning.

Contribution

It presents a novel structured multi-perspective reflection approach for LLMs, improving reasoning robustness and self-correction through prompt engineering without model retraining.

Findings

PR-CoT outperforms traditional CoT in accuracy and consistency.

Significant improvements in ethical decision-making tasks.

Ablation and human evaluations validate the effectiveness of multiple reflection perspectives.

Abstract

While Chain-of-Thought (CoT) prompting advances LLM reasoning, challenges persist in consistency, accuracy, and self-correction, especially for complex or ethically sensitive tasks. Existing single-dimensional reflection methods offer insufficient improvements. We propose MyGO Poly-Reflective Chain-of-Thought (PR-CoT), a novel methodology employing structured multi-perspective reflection. After initial CoT, PR-CoT guides the LLM to self-assess its reasoning across multiple predefined angles: logical consistency, information completeness, biases/ethics, and alternative solutions. Implemented purely via prompt engineering, this process refines the initial CoT into a more robust and accurate final answer without model retraining. Experiments across arithmetic, commonsense, ethical decision-making, and logical puzzles, using GPT-three point five and GPT-four models, demonstrate PR-CoT's superior performance. It significantly outperforms traditional CoT and existing reflection methods in logical consistency and error correction, with notable gains in nuanced domains like ethical decision-making. Ablation studies, human evaluations, and qualitative analyses further validate the contribution of each reflection perspective and the overall efficacy of our poly-reflective paradigm in fostering more reliable LLM reasoning.