Large Language Models Have Intrinsic Meta-Cognition, but Need a Good Lens

2025-06-10

AI Summary

This work addresses the underexplored problem of evaluating large language models’ (LLMs’) metacognitive ability—their capacity to self-detect errors at individual reasoning steps—beyond coarse-grained metrics like overall perplexity or post-hoc verification. We propose AutoMeco, the first automated framework for fine-grained, step-level metacognitive evaluation. To enable adaptive calibration of the metacognitive “lens,” we further introduce MIRA (Metacognitive Intrinsic Reward Adjustment), a training-free, Markov decision process–based intrinsic reward tuning strategy. Experiments across three mathematical reasoning benchmarks and three major LLM families demonstrate that AutoMeco is both principled and scalable; moreover, MIRA substantially improves step-level error detection accuracy and robustness. Collectively, this work establishes a novel paradigm for enhancing LLM reliability through rigorous, interpretable, and step-aware metacognitive assessment.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Search and Optimization: Metareasoning and Metaheuristics
• Natural Language Processing: (Large) Language Models

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Previous research has primarily focused on the cognitive error detection capabilities of Large Language Models (LLMs), often prompting them to analyze mistakes in reasoning chains. However, few studies have examined the meta-cognitive abilities of LLMs (e.g., their self-awareness of step errors), which are crucial for their reliability. While studies on LLM self-evaluation present some measures, such as perplexity, which can reflect the answer correctness and be viewed as the lens of meta-cognition, they lack step-level analysis and adaptation. This paper studies the evaluation of LLM meta-cognition using the current lenses and how to improve these lenses. Specifically, we propose AutoMeco, an Automated Meta-cognition Evaluation framework for benchmarking the existing lenses. Furthermore, a training-free Markovian Intrinsic Reward Adjustment strategy, MIRA, is proposed to boost current meta-cognition lenses. Experimental results on three mathematical reasoning datasets and three LLMs show the reasonableness of AutoMeco by comparing it with Best-of-N verification. Moreover, the meta-cognition ability of LLMs can be better evaluated using MIRA.

Problem

• Evaluating LLM meta-cognition using current lenses
• Improving meta-cognition lenses for better error detection
• Assessing step-level self-awareness in Large Language Models

Innovation

• AutoMeco framework for meta-cognition evaluation
• MIRA strategy to enhance meta-cognition lenses
• Step-level analysis with intrinsic reward adjustment