Investigating Tool-Memory Conflicts in Tool-Augmented LLMs

TL;DR

This paper proposes 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.

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.

Figures (3)

• Figure 1: Illustration of tool-memory Conflict (MCT). ✓ and ✗ mean the output is correct or wrong respectively compared to the ground truth answer.
• Figure 2: Tool-Memory Conflict across different domains.
• Figure 3: Tool-Memory Conflict across different types of tasks.