arXiv AI
Detecting and Repairing Hallucinations in Retrieval-Augmented Generation
Read the original on arXiv AI →The Flow has not summarised this story yet — read it at arXiv AI.
Related stories
arXiv AI
Oct 9
RAG-Stress: Probing the Limits of Evidence Reliance in Retrieval-Augmented Generation
The paper introduces RAG‑Stress, a diagnostic protocol that tests how retrieval‑augmented generation models handle misleading evidence. By fixing the question and reference answer, editing a single assertion to support an incorrect answer, and varying source‑priority policies and evidence positions, the study measures misleading rate (MR) and clean accuracy across fifteen systems on multiple QA datasets. Results show that prioritizing documents increases MR by 10.9–13.5 percentage points, with the highest MR when the misleading span appears at the end of the evidence.
arXiv AI
Sep 10
A Hierarchical Consistency Framework for Auditing Retrieval-Augmented Generation Systems
arXiv:2609.07075v1 Announce Type: new Abstract: Retrieval-augmented generation (RAG) is commonly evaluated by whether the final answer is correct. That test is insufficient: an answer can match its r...
arXiv Computation and Language
Sep 10
In RAG We Trust? Measuring Robustness of Retrieval-Augmented Generation Under Document Poisoning
arXiv:2609.09243v1 Announce Type: cross Abstract: Retrieval-augmented generation (RAG) grounds a language model in retrieved documents, which reduces hallucination but creates a new attack surface: i...
arXiv AI
Oct 5
Sentence-Level Context Sensitivity as a Training-Free Detector of Unsupported Content, Evaluated Against Trained Verifiers
The paper proposes a training‑free detector that uses sentence‑level context sensitivity to identify unsupported content in retrieval‑augmented generation (RAG) answers. By re‑scoring each sentence with full context, no context, and each chunk removed, the method flags the chunk whose removal most reduces a sentence’s likelihood as the likely source. Evaluated on RAGTruth, TofuEval, and RAGBench, the detector outperforms answer‑level faithfulness scores, achieving AUCs up to 0.745 and matching per‑chunk fact‑checkers while using only a fraction of the compute required by large‑language‑model judges.
arXiv Computation and Language
Oct 19
When Does Uncertainty Measure RAG Hallucination Risk?
The paper investigates whether uncertainty estimates can gauge hallucination risk in retrieval‑augmented generation (RAG). It finds that uncertainty is informative only when retrieved evidence constrains the answer: with good evidence, unsupported claims are less likely, whereas poor retrieval coverage lets the model rely on internal knowledge and assign high probability to unsupported claims. Experiments on HotpotQA and SQuAD across four instruction‑tuned models show that perplexity‑based uncertainty achieves AUROC scores of 0.81–0.91 for detecting targeted unsupported edits, while length‑only classifiers perform near chance. The study also demonstrates that adding a second passage can recover detection performance and that combining uncertainty with a semantic‑drift score further improves calibration.
arXiv Computation and Language
Sep 28
Stale-Document Poisoning: When Outdated Retrieval Overrides Correct Model Answers
The paper introduces "stale‑document poisoning," a temporal alignment failure where outdated retrieval evidence causes retrieval‑augmented generation models to produce incorrect answers even when the model could answer correctly without retrieval. A benchmark of 317 verified knowledge reversals across medicine, law, software, and platform policy shows that outdated evidence flips 30–37% of Llama and Qwen answers, rising to 66–75% when models are explicitly instructed to trust the document. The study demonstrates that providing explicit validity information and a recency‑aware re‑ranker can substantially reduce poisoning, highlighting the need for models to assess whether retrieved evidence remains applicable.