DRAGged into Conflicts: Detecting and Addressing Conflicting Sources in Search-Augmented LLMs

TL;DR

This paper introduces a new taxonomy and benchmark for detecting and resolving conflicting information in retrieval-augmented large language models, highlighting current challenges and potential improvements.

Contribution

It proposes a novel taxonomy of knowledge conflicts, creates the CONFLICTS benchmark with expert annotations, and evaluates LLMs' ability to handle conflicting sources.

Findings

LLMs often struggle to resolve conflicts effectively.

Explicit reasoning about conflicts improves response quality.

Significant room for future improvements in conflict resolution.

Abstract

Retrieval Augmented Generation (RAG) is a commonly used approach for enhancing large language models (LLMs) with relevant and up-to-date information. However, the retrieved sources can often contain conflicting information and it remains unclear how models should address such discrepancies. In this work, we first propose a novel taxonomy of knowledge conflict types in RAG, along with the desired model behavior for each type. We then introduce CONFLICTS, a high-quality benchmark with expert annotations of conflict types in a realistic RAG setting. CONFLICTS is the first benchmark that enables tracking progress on how models address a wide range of knowledge conflicts. We conduct extensive experiments on this benchmark, showing that LLMs often struggle to appropriately resolve conflicts between sources. While prompting LLMs to explicitly reason about the potential conflict in the retrieved documents significantly improves the quality and appropriateness of their responses, substantial room for improvement in future research remains.