Investigating Tool-Memory Conflicts in Tool-Augmented LLMs

2026-01-14

AI Summary

This work addresses a critical yet previously unexamined issue in tool-augmented large language models (LLMs): performance degradation in STEM tasks due to conflicts between the model’s internal parametric knowledge and externally retrieved tool knowledge. The authors formally introduce and define this phenomenon as “Tool-Memory Conflict” (TMC) and systematically investigate how LLMs prioritize between these two knowledge sources across diverse scenarios. Through comprehensive evaluation of prevailing mitigation strategies—including prompt engineering and retrieval-augmented generation (RAG)—the study demonstrates that current approaches consistently fail to resolve TMC, with particularly pronounced performance deterioration in high-precision STEM contexts. These findings underscore the pervasiveness and severity of TMC, offering crucial insights for the future design of knowledge coordination mechanisms in tool-augmented language models.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Tool-augmented large language models (LLMs) have powered many applications. However, they are likely to suffer from knowledge conflict. In this paper, we propose a new type of knowledge conflict -- Tool-Memory Conflict (TMC), where the internal parametric knowledge contradicts with the external tool knowledge for tool-augmented LLMs. We find that existing LLMs, though powerful, suffer from TMC, especially on STEM-related tasks. We also uncover that under different conditions, tool knowledge and parametric knowledge may be prioritized differently. We then evaluate existing conflict resolving techniques, including prompting-based and RAG-based methods. Results show that none of these approaches can effectively resolve tool-memory conflicts.

Problem

• Tool-Memory Conflict
• tool-augmented LLMs
• knowledge conflict
• parametric knowledge
• external tool knowledge

Innovation

• Tool-Memory Conflict
• tool-augmented LLMs
• knowledge conflict
• parametric knowledge
• external tool knowledge