Language Models Coupled with Metacognition Can Outperform Reasoning Models

2025-08-25

AI Summary

Large language models (LLMs) suffer from insufficient logical rigor, while large reasoning models (LRMs) incur high computational overhead and slow inference—posing a fundamental trade-off between accuracy and efficiency. Method: This paper proposes SOFAI-LM, a cognitive architecture integrating a lightweight metacognitive module that enables dynamic, zero-shot collaboration between LLMs and LRMs. The module monitors reasoning processes in real time, detects logical weaknesses, and selectively activates the LRM for targeted intervention; it further employs example-driven feedback and problem-domain-adaptive information propagation to ensure reasoning consistency. Contribution/Results: On graph coloring and code debugging tasks, SOFAI-LM matches or exceeds the solution accuracy of standalone LRMs while reducing inference latency by 42%–68%, thereby achieving a significant improvement in the joint optimization of efficiency and correctness for complex reasoning tasks.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Search and Optimization: Metareasoning and Metaheuristics

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 language models (LLMs) excel in speed and adaptability across various reasoning tasks, but they often struggle when strict logic or constraint enforcement is required. In contrast, Large Reasoning Models (LRMs) are specifically designed for complex, step-by-step reasoning, although they come with significant computational costs and slower inference times. To address these trade-offs, we employ and generalize the SOFAI (Slow and Fast AI) cognitive architecture into SOFAI-LM, which coordinates a fast LLM with a slower but more powerful LRM through metacognition. The metacognitive module actively monitors the LLM's performance and provides targeted, iterative feedback with relevant examples. This enables the LLM to progressively refine its solutions without requiring the need for additional model fine-tuning. Extensive experiments on graph coloring and code debugging problems demonstrate that our feedback-driven approach significantly enhances the problem-solving capabilities of the LLM. In many instances, it achieves performance levels that match or even exceed those of standalone LRMs while requiring considerably less time. Additionally, when the LLM and feedback mechanism alone are insufficient, we engage the LRM by providing appropriate information collected during the LLM's feedback loop, tailored to the specific characteristics of the problem domain and leads to improved overall performance. Evaluations on two contrasting domains: graph coloring, requiring globally consistent solutions, and code debugging, demanding localized fixes, demonstrate that SOFAI-LM enables LLMs to match or outperform standalone LRMs in accuracy while maintaining significantly lower inference time.

Problem

• Combining fast LLMs with metacognition to enhance reasoning without fine-tuning
• Improving LLM problem-solving on logic tasks like graph coloring and debugging
• Reducing computational costs while matching or exceeding standalone reasoning models

Innovation

• Combines fast LLM with slow LRM via metacognition
• Uses iterative feedback to refine solutions without fine-tuning
• Engages LRM only when LLM feedback is insufficient