Research map / Large language models / Language, safety and interpretability
Zero-source LLM Hallucination Detection with Human-like Criteria Probing
Jiahao Yang, Shuhai Zhang, Hailong Kang, Feng Liu, Qi Chen, Mingkui Tan
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 |
|---|---|---|---|
| add GRPO alignment training with weak labels (ablation) | HCPD · TriviaQA (Llama-3.1-8b) | AUROC: 66.54 -> 86.25 | Improved |
| add criteria probing to plain self-evaluation (pre-RL) (ablation) | self-evaluation · TriviaQA (Llama-3.1-8b) | AUROC: 56.07 -> 66.54 | Improved |
| use graded score-alignment reward instead of binary reward (ablation) | HCPD · TriviaQA (Llama-3.1-8b) | AUROC: 79.06 -> 86.25 | Improved |
| use graded score-alignment reward instead of binary reward (ablation) | HCPD · CoQA (Llama-3.1-8b) | AUROC: 51.75 -> 90.07 | Improved |
| increase sampling size K from 1 to 5 (ablation) | HCPD · TriviaQA (Llama-3.1-8b) | AUROC: 85.21 -> 86.25 | Improved |
| increase sampling size K from 1 to 5 (ablation) | HCPD · NQ Open (Llama-3.1-8b) | AUROC: 86.89 -> 90.38 | Improved |
| increase sampling size K from 5 to 10 (ablation) | HCPD · CoQA (Llama-3.1-8b) | AUROC: 90.07 -> 89.85 | No clear effect |
| increase sampling size K from 1 to 5 (ablation) | HCPD · inference time | seconds: 0.2349 -> 1.1313 | Hurt |
| use ROUGE-based reward instead of BLEURT | HCPD · TriviaQA (Llama-3.1-8b) | AUROC: 89.17 under ROUGE labels vs best baseline SAPLMA 87.22 | Improved |
| use DeepSeek-V3 judge as reward signal | HCPD · TriviaQA (Llama-3.1-8b) | AUROC: 80.42 vs SAPLMA 79.99 | Improved |
| apply detector trained on Llama-3.1-8b to Qwen-3-8b outputs | HCPD vs TSV · TriviaQA transfer | AUROC: HCPD 78.89 vs TSV 59.89 | 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 2 of 2 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
- Scalable Token-Level Hallucination Detection in Large Language Models (NeurIPS 2026 · similarity 0.93)
- LLM-Check: Investigating Detection of Hallucinations in Large Language Models (NeurIPS 2024 · similarity 0.93)
- Joint Evaluation of Answer and Reasoning Consistency for Hallucination Detection in Large Reasoning Models (AAAI 2026 · similarity 0.91)
- ANAH-v2: Scaling Analytical Hallucination Annotation of Large Language Models (NeurIPS 2024 · similarity 0.91)
- HaDeMiF: Hallucination Detection and Mitigation in Large Language Models (ICLR 2025 · similarity 0.91)
- FactTest: Factuality Testing in Large Language Models with Finite-Sample and Distribution-Free Guarantees (ICML 2025 · 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.
