Decoding the Critique Mechanism in Large Reasoning Models

2026-03-17

AI Summary

This study investigates how large reasoning models (LRMs) detect and correct errors in intermediate reasoning steps, with a particular focus on the phenomenon where incorrect intermediate steps nonetheless yield correct final answers. By deliberately injecting arithmetic errors into chain-of-thought reasoning and combining feature-space analysis with latent representation manipulation, the work uncovers and formally names an intrinsic “latent critical ability” within LRMs. It further identifies an interpretable critical vector that encodes this capability. Remarkably, this vector enhances the model’s error detection and self-correction performance at test time without requiring additional training. Extensive experiments across multiple model scales and families demonstrate the effectiveness of the approach, yielding significant improvements in both reasoning accuracy and robustness.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Knowledge Representation and Reasoning: Common-Sense Reasoning
• Computer Vision: Large Vision Models

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Large Reasoning Models (LRMs) exhibit backtracking and self-verification mechanisms that enable them to revise intermediate steps and reach correct solutions, yielding strong performance on complex logical benchmarks. We hypothesize that such behaviors are beneficial only when the model has sufficiently strong "critique" ability to detect its own mistakes. This work systematically investigates how current LRMs recover from errors by inserting arithmetic mistakes in their intermediate reasoning steps. Notably, we discover a peculiar yet important phenomenon: despite the error propagating through the chain-of-thought (CoT), resulting in an incorrect intermediate conclusion, the model still reaches the correct final answer. This recovery implies that the model must possess an internal mechanism to detect errors and trigger self-correction, which we refer to as the hidden critique ability. Building on feature space analysis, we identify a highly interpretable critique vector representing this behavior. Extensive experiments across multiple model scales and families demonstrate that steering latent representations with this vector improves the model's error detection capability and enhances the performance of test-time scaling at no extra training cost. Our findings provide a valuable understanding of LRMs' critique behavior, suggesting a promising direction to control and improve their self-verification mechanism. Our code is available at https://github.com/mail-research/lrm-critique-vectors.

Problem

• Large Reasoning Models
• critique mechanism
• error recovery
• self-verification
• chain-of-thought

Innovation

• critique mechanism
• large reasoning models
• self-correction
• critique vector
• test-time scaling