小模型在误导性上下文仍会答错，提示拒答无效。
Prompt-Based Abstention Fails Under Misleading Context: A Controlled Study of Small Frozen RAG Models
- 设计配对测试集，对比四种上下文条件下的拒答表现。
- 误导性上下文下模型仍答对41.6%，63%直接复制错误实体。
- 现有验证机制难兼顾安全与覆盖，提示拒答本质不判真伪。
检索增强生成（RAG）中，缺失证据与误导性证据并非同一问题，但基于提示的拒答策略将其等同处理。模型仅在无上下文时拒答，面对误导性信息仍会作答。本文提出GRAB-RAG（分级拒答基准），在Natural Questions和HotpotQA上测试同一问题在支持、退化、缺失、误导四种上下文条件下的表现。误导情境下，将真实答案替换为错误答案并插入检索结果中。在三个小型冻结模型（3.8B–8B）上测试五种拒答策略。模型在证据缺失时拒答可靠，但在明确拒答提示下仍回答41.6%的误导问题，其中63%的答案直接复现植入的错误实体。思维链提示几乎无改善作用。生成侧冲突检测可将拒答率降至13.3%，但误删大量正确答案；自然语言推理验证器虽恢复覆盖率，但在参数记忆与误导段落一致时失效。提示拒答关注上下文是否充分，而非是否正确。两种验证器均无法在不牺牲覆盖率的前提下填补此差距。
原文摘要 · Abstract (English)
Missing and misleading evidence are not the same problem in retrieval-augmented generation (RAG), but prompt-based abstention treats them alike. Models abstain when context is absent, not when it is misleading. We introduce GRAB-RAG (Graded Abstention Benchmark for Retrieval-Augmented Generation), a paired benchmark that tests the same questions across four context conditions (supportive, degraded, missing, and misleading) in Natural Questions and HotpotQA. In the misleading condition, we edit a gold passage to support a wrong answer and place it among other retrieved passages. We test five abstention policies on three small frozen models (3.8B--8B) across two QA benchmarks. Models abstain reliably when evidence is missing, but under explicit abstention prompting still answer 41.6% of misleading questions, with 63% of those answers echoing the planted wrong entity verbatim. Chain-of-thought provides little additional benefit. A generator-side conflict check cuts the rate to 13.3% but discards many correct answers, while an NLI verifier recovers that coverage but fails when parametric memory and the misleading passage agree on the same wrong answer. Prompt-based abstention asks whether context is sufficient, not whether it is correct. Neither verifier closes this gap without trading coverage for safety.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。