Research map / Large language models / Retrieval and knowledge
Can LLMs Refuse Questions They Do Not Know? Measuring Knowledge-Aware Refusal in Factual Tasks
Wenbo Pan, Jie Xu, Qiguang Chen, Junhao Dong, Libo Qin, Xinfeng Li, Yu Haining, Xiaohua Jia
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| prompt models to refuse more often (ablation) | LLMs · SimpleQA | Refusal Index: avg normalized difference +0.07 | No clear effect |
| prompt models to refuse more often (ablation) | LLMs · SimpleQA | C/A: avg normalized difference +0.49 | Improved |
| prompt models to refuse more often (ablation) | LLMs · SimpleQA | Correct answer rate: avg normalized difference -0.80 | Hurt |
| remove accuracy and refusal rate effects from ranking (ablation) | F-score metric · 8 evaluation settings | Kendall's W: 0.90 -> 0.10 (remove correct rate) | Hurt |
| remove accuracy and refusal rate effects from ranking (ablation) | Refusal Index · 8 evaluation settings | Kendall's W: 0.47 -> 0.49 (both removed) | No clear effect |
| use higher model accuracy | LLMs · SimpleQA | Refusal Index: R^2=0.242 with correct answer rate | No clear effect |
| use Refusal Index as measure of refusal vs AUROC with P(Answering) | Refusal Index · SimpleQA | Correlation with AUROC: 85% | Improved |
| remove ground truth from the context | LLMs · FaithEval-based hallucination benchmarks | Refusal Index: 0.48 -> 0.24 (PreciseWiki vs Inconsistency avg) | Hurt |
| larger model scale | LLMs · SimpleQA | Refusal Index | No clear effect |
Closest papers by meaning
- Utilize the Flow Before Stepping into the Same River Twice: Certainty Represented Knowledge Flow for Refusal-Aware Instruction Tuning (AAAI 2025 · similarity 0.91)
- OR-Bench: An Over-Refusal Benchmark for Large Language Models (ICML 2025 · similarity 0.88)
- Latent-space Attacks for Refusal Evasion in Language Models (NeurIPS 2026 · similarity 0.87)
- The Geometry of Refusal in Large Language Models: Concept Cones and Representational Independence (ICML 2025 · similarity 0.87)
- AbstentionBench: Reasoning LLMs Fail on Unanswerable Questions (NeurIPS 2025 · similarity 0.86)
- SORRY-Bench: Systematically Evaluating Large Language Model Safety Refusal (ICLR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
