Probing Latent Knowledge Conflict for Faithful Retrieval-Augmented Generation

Linfeng Gao, Baolong Bi, Zheng Yuan, Le Wang, Zerui Chen, Zhimin Wei, Shenghua Liu, Qinggang Zhang, Jinsong Su

Keywords: RAG, Faithfulness, Knowledge Conflict, LLM

Abstract: Retrieval-Augmented Generation (RAG) has emerged as a powerful paradigm to enhance the factuality of Large Language Models (LLMs). However, existing RAG systems often suffer from an unfaithfulness issue, where the model's response contradicts evidence from the retrieved context. Existing approaches to improving contextual faithfulness largely rely on external interventions, such as prompt engineering, decoding constraints, or reward-based fine-tuning. These works treat the LLM as a black box and overlook a crucial question: how does the LLM internally integrate retrieved evidence with its parametric memory, particularly under knowledge conflicts? To address this gap, we conduct a probing-based analysis of hidden-state representations in LLMs and observe three findings: knowledge integration occurs hierarchically, conflicts manifest as latent signals at the sentence level, and irrelevant context is often amplified when aligned with parametric knowledge. Building on these findings, we propose **CLEAR** (**C**onflict-**L**ocalized and **E**nhanced **A**ttention for **R**AG), a framework that (i) decomposes context into fine-grained sentence-level knowledge, (ii) employs hidden-state probing to localize conflicting knowledge, and (iii) introduces conflict-aware fine-tuning to guide the model to accurately integrate retrieved evidence. Extensive experiments across three benchmarks demonstrate that CLEAR substantially improves both accuracy and contextual faithfulness, consistently outperforming strong baselines under diverse conflict conditions. The related resources are available at https://anonymous.4open.science/r/CLEAR-CF6B.

Primary Area: applications to computer vision, audio, language, and other modalities

Submission Number: 22987