Trust or Abstain? A Self-Aware RAG Approach

2026-05-11

AI Summary

This work addresses the unreliability of retrieval-augmented generation (RAG) systems when parametric and retrieved knowledge conflict, a problem exacerbated by the lack of self-awareness in large language models (LLMs). To this end, we introduce, for the first time, LLM self-knowledge into the RAG decision process. We construct a model-specific knowledge conflict benchmark and propose SABER, a fine-tuning-free Self-Aware Belief Estimator that explicitly evaluates source credibility through multi-trajectory reasoning, self-prior modeling, and a lightweight reliability predictor. SABER enables a risk-coverage controllable abstention mechanism, significantly improving end-to-end accuracy and faithfulness under knowledge conflicts across four mainstream LLMs. Its abstention performance achieves Pareto superiority over all existing prompt-based methods.

Abstract

Retrieval-augmented generation (RAG) improves large language models (LLMs) by incorporating external evidence, but it also introduces knowledge conflicts when retrieved contextual knowledge (CK) and parametric knowledge (PK) disagree or are both unreliable. Existing approaches mainly coordinate which source to use, without explicitly asking whether each answer path is correct. We argue that faithful RAG requires LLM self-awareness, namely the ability to recognize the limits of its own knowledge and reasoning. To ground this problem, we construct a model-specific, ground-truth-aligned knowledge-conflict benchmark by evaluating LLM backbones on PK-only and CK-conditioned answer paths over approximately 69K query-context instances per backbone, drawn from five conflict-QA datasets. We then introduce SABER, a Self-Aware Belief Estimator for RAG that requires no LLM fine-tuning. SABER combines a self-prior with PK-side and CK-side conditional reasoning representations from multi-trace inference, then estimates reliability beliefs with two lightweight predictors to drive a 4-cell decision over trust PK, trust CK, trust either, or abstain. Across four LLM backbones, SABER improves end-to-end accuracy and conflict-specific faithfulness over ten inference-time and fine-tuning baselines, with the largest gains on conflict-heavy datasets. Under abstention, SABER's risk-coverage curve Pareto-dominates every prompt-based abstainer, providing a tunable balance between coverage and answer risk. Our code is available at https://github.com/xizhu1022/SABER.

Problem

• Retrieval-Augmented Generation
• Knowledge Conflict
• Self-Awareness
• Abstention
• Trust Calibration

Innovation

• Self-Aware RAG
• Knowledge Conflict
• Abstention Mechanism
• Retrieval-Augmented Generation
• Faithful Reasoning