Research map / Large language models / Agents, code and math
HalluGuard: Demystifying Data-Driven and Reasoning-Driven Hallucinations in LLMs
Xinyue Zeng, Junhong Lin, Yujun Yan, Feng Guo, Liang Shi, Jun Wu, Dawei Zhou
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 HalluGuard score instead of Inside | detector · RAGTruth (QwQ-32B) | AUROC: 77.72 -> 84.59 | Improved |
| use HalluGuard score instead of Inside | detector · MATH-500 (QwQ-32B) | AUROC: 80.8 -> 81.76 | Improved |
| use HalluGuard score instead of RACE/Inside | detector · TruthfulQA (Mistral-7B) | AUROC: 71.06 -> 77.05 | Improved |
| use HalluGuard on smaller backbone | detector · HaluEval (Llama2-7B) | AUPRC: 72.89 | Improved |
| use HalluGuard on Llama2-13B | detector · GSM8K | AUROC: 75.79 -> 79.01 | Improved |
| use HalluGuard on Llama2-70B | detector · SQuAD | AUROC: 81.24 -> 83.8 | Improved |
| use HalluGuard on Llama2-70B | detector · HaluEval | AUROC: 74.62 -> 80.79 | Improved |
| guide beam search with HalluGuard score instead of IO Prompt | Qwen2.5-Math-7B · MATH-500 | accuracy: 72.70 -> 81.00 | Improved |
| guide beam search with HalluGuard score instead of IO Prompt | Llama3.1-8B · Natural | accuracy: 55.24 -> 70.96 | Improved |
| use HalluGuard for fine-grained semantic hallucination | detector · PAWS (QwQ-32B) | AUROC: 91.24 | Improved |
| use HalluGuard on GPT-2 | detector · PAWS | AUROC: 83.27 | Improved |
| add the -log kappa^2 stability term (ablation) | HalluGuard score · MATH-500 (increasing reasoning drift) | AUROC | Improved |
Closest papers by meaning
- Scalable Token-Level Hallucination Detection in Large Language Models (NeurIPS 2026 · similarity 0.94)
- LLM-Check: Investigating Detection of Hallucinations in Large Language Models (NeurIPS 2024 · similarity 0.94)
- HaDeMiF: Hallucination Detection and Mitigation in Large Language Models (ICLR 2025 · similarity 0.93)
- Auditing Meta-Cognitive Hallucinations in Reasoning Large Language Models (NeurIPS 2025 · similarity 0.93)
- ReDeEP: Detecting Hallucination in Retrieval-Augmented Generation via Mechanistic Interpretability (ICLR 2025 · similarity 0.92)
- MIRAGE: Assessing Hallucination in Multimodal Reasoning Chains of MLLM (NeurIPS 2025 · 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.
