Research map / Large language models / Language, safety and interpretability
Attention Sinks as Internal Signals for Hallucination Detection in Large Language Models
Jakub Binkowski, Kamil Adamczewski, Tomasz Jan Kajdanowicz
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 |
|---|---|---|---|
| use SinkProbe sink-score features instead of prior attention-based detectors | SinkProbe · 7 hallucination benchmarks x 4 LLMs | ROC-AUC: best in 23 of 28 model-dataset pairs | Improved |
| SinkProbe vs best baseline on UMWP with Llama3.2-3B | SinkProbe · UMWP, Llama3.2-3B | ROC-AUC: 0.809 AttnLogDet vs 0.757 | Hurt |
| SinkProbe vs best baseline on GSM8K with Mistral-Nemo | SinkProbe · GSM8K, Mistral-Nemo | ROC-AUC: 0.841 LookbackLens vs 0.811 | Hurt |
| SinkProbe vs best baseline on TruthfulQA with Phi3.5 | SinkProbe · TruthfulQA, Phi3.5 | ROC-AUC: 0.797 MTopDiv vs 0.761 | Hurt |
| SinkProbe vs best baseline on UMWP with Llama3.1-8B | SinkProbe · UMWP, Llama3.1-8B | ROC-AUC: 0.838 -> 0.879 | Improved |
| increase top-k sink scores per head from 1 to between 2 and 5 (ablation) | SinkProbe · GSM8K, UMWP, NQ-Open, TruthfulQA (Llama3.2-3B) | ROC-AUC | Improved |
| increase k beyond small values (k>=10) (ablation) | SinkProbe · GSM8K, UMWP, NQ-Open, TruthfulQA (Llama3.2-3B) | ROC-AUC | No clear effect |
| use sink scores instead of raw attention eigenvalues for top-k features | SinkProbe · GSM8K, UMWP, NQ-Open, TruthfulQA (Llama3.2-3B) | ROC-AUC | Improved |
| train probe with L1 regularization (ablation) | SinkProbe · GSM8K, NQ-Open, TruthfulQA, UMWP | ROC-AUC: retains 1% to 4% of features | No clear effect |
Closest papers by meaning
- Scalable Token-Level Hallucination Detection in Large Language Models (NeurIPS 2026 · similarity 0.93)
- Entropy Distribution as a Fingerprint for Hallucinations in Generative Models (NeurIPS 2026 · similarity 0.92)
- HaDeMiF: Hallucination Detection and Mitigation in Large Language Models (ICLR 2025 · similarity 0.92)
- PAS: Prelim Attention Score for Detecting Object Hallucinations in Large Vision-Language Models (CVPR 2026 · similarity 0.92)
- Detecting and Mitigating Hallucination in Large Vision Language Models via Fine-Grained AI Feedback (AAAI 2025 · similarity 0.92)
- LLM-Check: Investigating Detection of Hallucinations in Large Language Models (NeurIPS 2024 · similarity 0.92)
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.
