Wrong-of-Thought: An Integrated Reasoning Framework with Multi-Perspective Verification and Wrong Information

TL;DR

Wrong-of-Thought (WoT) is proposed, which includes two core modules: Multi-Perspective Verification: A multi-perspective verification method for accurately refining the reasoning process and result, and Wrong Information Utilization: Utilizing wrong information to alert LLMs and reduce the probability of LLMs making same mistakes.

Abstract

Chain-of-Thought (CoT) has become a vital technique for enhancing the performance of Large Language Models (LLMs), attracting increasing attention from researchers. One stream of approaches focuses on the iterative enhancement of LLMs by continuously verifying and refining their reasoning outputs for desired quality. Despite its impressive results, this paradigm faces two critical issues: (1) Simple verification methods: The current paradigm relies solely on a single verification method. (2) Wrong Information Ignorance: Traditional paradigms directly ignore wrong information during reasoning and refine the logic paths from scratch each time. To address these challenges, we propose Wrong-of-Thought (WoT), which includes two core modules: (1) Multi-Perspective Verification: A multi-perspective verification method for accurately refining the reasoning process and result, and (2) Wrong Information Utilization: Utilizing wrong information to alert LLMs and reduce the probability of LLMs making same mistakes. Experiments on 8 popular datasets and 5 LLMs demonstrate that WoT surpasses all previous baselines. In addition, WoT exhibits powerful capabilities in difficult computation tasks.

Figures (8)

• Figure 1: Previous multi-thoughts integration methods (a) vs. Wrong-of-Thought (b). Previous methods only used a Single Verification and did not utilize the wrong information. In contrast, WoT offers Multi-Perspective Verification and utilizes Wrong Information.
• Figure 2: XoT Framework. First, select a reasoning method, either PoT or EoT and then apply assertion verification to make a judgment. If the reasoning is found to be incorrect, switch to the alternative method and restart the reasoning. Verify again, and if the verification is correct, return the answer. If the reasoning reaches the third step, utilize CoT reasoning as the answer.
• Figure 3: Overview of the Wrong-of-Thought (WoT) framework, incorporating three core modules: Planning and Solving ($\S \ref{['plan']}$), Multi-Perspective Verification ($\S \ref{['MPV']}$), and Wrong Information Utilization ($\S \ref{['WRA']}$).
• Figure 4: Performance comparison results from various verification perspectives. "Voting" represents the final judgment after voting from the three perspectives.
• Figure 5: Comparison of performance without utilizing wrong reasoning information and with integrated wrong reasoning information.