Research map / Large language models / Language, safety and interpretability
Bolster Hallucination Detection via Prompt-Guided Data Augmentation
Wenyun Li, Zheng Zhang, Dongmei Jiang, Xiangyuan Lan
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 prompt-guided augmented data with CM score instead of HaloScope | LLaMA-3.1-7B · TruthfulQA | AUROC (%): 70.16 -> 73.20 | Improved |
| use PALE instead of HaloScope | LLaMA-3.1-7B · TyDi QA-GP | AUROC (%): 78.38 -> 85.34 | Improved |
| use PALE instead of Lexical Similarity | LLaMA-3.1-7B · CoQA | AUROC (%): 74.36 -> 57.94 | Hurt |
| use PALE instead of HaloScope | Qwen-2.5-7B · TruthfulQA | AUROC (%): 73.42 -> 83.68 | Improved |
| use PALE instead of HaloScope | OPT-6.7B · CoQA | AUROC (%): 67.24 -> 78.64 | Improved |
| scale base LLM to 13B/14B | PALE · TruthfulQA | AUROC (%): 73.20 -> 75.51 | Improved |
| train on both truthful and hallucinated data instead of truthful only (ablation) | PALE · TruthfulQA | AUROC gain: 11.7% | Improved |
| train on both truthful and hallucinated data instead of hallucinated only (ablation) | PALE · TruthfulQA | AUROC gain: 39.0% | Improved |
| use different LLMs for augmentation (Claude best) (ablation) | PALE · TruthfulQA | AUROC | No clear effect |
| vary augmentation prompt template (10 templates) (ablation) | PALE · TruthfulQA | AUROC | No clear effect |
| use CM score instead of direct binary classifier (ablation) | PALE · Four datasets | AUROC | Improved |
| extract embeddings from middle layer instead of high layers (ablation) | PALE · TruthfulQA | AUROC | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- HaloScope — HaloScope: Harnessing Unlabeled LLM Generations for Hallucination Detection (NeurIPS 2024) · better on 20 of 20 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Sample Transform Cost-Based Training-Free Hallucination Detector for Large Language Models (NeurIPS 2026 · similarity 0.92)
- Scalable Token-Level Hallucination Detection in Large Language Models (NeurIPS 2026 · similarity 0.92)
- LLM-Check: Investigating Detection of Hallucinations in Large Language Models (NeurIPS 2024 · similarity 0.92)
- Steer LLM Latents for Hallucination Detection (ICML 2025 · similarity 0.92)
- HaDeMiF: Hallucination Detection and Mitigation in Large Language Models (ICLR 2025 · similarity 0.91)
- Detecting and Preventing Hallucinations in Large Vision Language Models (AAAI 2024 · similarity 0.91)
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.
