Elamur Research⌕
☰
home / papers / 2026 / Identifying the Achilles' Heel: An Iterative Method for Uncovering Factual Errors in Large Language Models

Identifying the Achilles' Heel: An Iterative Method for Uncovering Factual Errors in Large Language Models

Paper · 2026 · Wenxuan Wang, Yuk-Kit Chan, Zixuan Ling et al.

Summary

Large language models often state false facts with confidence. Existing factuality benchmarks have four main weaknesses: they need costly manual annotation, their static test sets may have leaked into training data, they cover few topics or question formats, and they cannot steer testing toward a model's weak spots. The authors present HalluHunter, a fully automatic testing framework. It pulls fact triplets from a knowledge base through SPARQL queries and arranges them as a directed graph. Rule-based NLP, including part-of-speech tagging and NER, then turns the triplets into Yes/No, multiple-choice and WH questions that need either one or several reasoning hops. The first round samples triplets at random. Later rounds favour relations where the model scored poorly and entities resembling those it got wrong. Across nine commercial and open LLMs, up to 55% of the questions produced factual errors, and accuracy dropped steadily with each adaptive round. Because questions are generated fresh at test time and the method works on any topic, it reduces contamination risk and makes factuality testing cheaper.

Key points

  • The knowledge source is Wikidata, which holds over 100 million items, and users can narrow it by topic, for example "occupation: emperor." Multiple-choice distractors are taken from other graph edges that share the same relation.

  • The nine models tested were GPT-3.5-Turbo, GPT-4-Turbo, GPT-4o, DeepSeek-v3, Claude-Sonnet-4, Claude-3.5-Haiku, Gemini-2.0-Flash, Qwen-3 and Qwen-3-Reasoning.

  • From the seed round to Trial 5, median accuracy fell as follows:

    • Humanities: 0.712 to 0.462 (−32.7%)
    • Social Science: 0.699 to 0.384 (−40.2%)
    • STEM: 0.649 to 0.373 (−41.8%)

    The decline also appeared on 2-hop questions.

  • The adaptive selection did not make the questions less diverse. Over 5,000 questions, average knowledge-graph coverage was 0.473 for Trial 5 versus 0.406 for random generation.

  • The authors note two limitations: results depend on the quality of the knowledge graph, and the paper does not propose any way to fix factual errors. Code and data are released at github.com/Mysterchan/HalluHunter.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Johns Hopkins UniversityBaltimore · United States240 works · 96,487 citations🏢TencentShenzhen · China134 works · 19,743 citations🎓Nanyang Technological UniversitySingapore · Singapore156 works · 25,644 citations🎓Renmin University of ChinaBeijing · China114 works · 16,118 citations🎓Chinese University of Hong KongHong Kong · China126 works · 14,073 citations🎓Chinese University of Hong Kong, ShenzhenShenzhen · China47 works · 2,596 citations🏢Xiaohongshu (Little Red Book)Shanghai · China11 works · 107 citations

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksSocial Sciences · 1,772 worksSociology and Political Science · 577 works

🧭 Topics 1Research subjects and languages the work is about.Language models & pretraining · 3,320 works

Cite

Cite as Wenxuan Wang, Yuk-Kit Chan, Zixuan Ling, Shi Juluan, Youliang Yuan, Jen-tse Huang, Yifei Zhang, Wenxiang Jiao, Zhaopeng Tu and Michael R. Lyu. 2026. Identifying the Achilles' Heel: An Iterative Method for Uncovering Factual Errors in Large Language Models. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.1714

@inproceedings{wang2026identifying,
  title = {{Identifying the Achilles' Heel: An Iterative Method for Uncovering Factual Errors in Large Language Models}},
  author = {Wenxuan Wang, Yuk-Kit Chan, Zixuan Ling, Shi Juluan, Youliang Yuan, Jen-tse Huang, Yifei Zhang, Wenxiang Jiao, Zhaopeng Tu and Michael R. Lyu},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1714},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1714}
}

References

39 references, 16 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Rachith Aiyappa, Jisun An, Haewoon Kwak, and Yong-Yeol Ahn. 2023. Can we trust the evaluation on chatgpt? ArXiv, abs/2303.12767.
  2. Mehdi Ali, Max Berrendorf, Charles Tapley Hoyt, Laurent Vermue, Sahand Sharifzadeh, Volker Tresp, and Jens Lehmann. 2021. PyKEEN 1.0: A Python Library for Training and Evaluating Knowledge Graph Embeddings. Journal of Machine Learning Research, 22(82):1–6. find
  3. Yejin Bang, Samuel Cahyawijaya, Nayeon Lee, Wenliang Dai, Dan Su, Bryan Wilie, Holy Lovenia, Ziwei Ji, Tiezheng Yu, Willy Chung, Quyet V. Do, Yan Xu, and Pascale Fung. 2023. A multitask, multilingual, multimodal evaluation of chatgpt on reasoning, hallucination, and interactivity. ArXiv, abs/2302.04…
  4. Shulin Cao, Jiaxin Shi, Liangming Pan, Lunyiu Nie, Yutong Xiang, Lei Hou, Juanzi Li, Bin He, and Hanwang Zhang. 2022. KQA pro: A dataset with explicit compositional programs for complex question answering over knowledge base. In Proceedings of the 60th Annual Meeting of the Association for Computati… find
  5. Songqiang Chen, Shuo Jin, and Xiaoyuan Xie. 2021. Testing your question answering software via asking recursively. 2021 36th IEEE/ACM International Conference on Automated Software Engineering (ASE), pages 104–116. find
  6. Yulong Chen, Yang Liu, Jianhao Yan, Xuefeng Bai, Ming Zhong, Yinghao Yang, Ziyi Yang, Chenguang Zhu, and Yue Zhang. 2024. See what llms cannot answer: A self-challenge framework for uncovering llm weaknesses. Preprint, arXiv:2408.08978. — [arXiv:2408.08978](https://arxiv.org/abs/2408.08978) find
  7. Jiale Cheng, Yida Lu, Xiaotao Gu, Pei Ke, Xiao Liu, Yuxiao Dong, Hongning Wang, Jie Tang, and Minlie Huang. 2024. Autodetect: Towards a unified framework for automated weakness detection in large language models. Preprint, arXiv:2406.16714. find
  8. Christopher Clark, Kenton Lee, Ming-Wei Chang, Tom Kwiatkowski, Michael Collins, and Kristina Toutanova. 2019. Boolq: Exploring the surprising difficulty of natural yes/no questions. In North American Chapter of the Association for Computational Linguistics.
  9. Yue Zhang Cunxiang Wang, Pai. 2021. Can generative pre-trained language models serve as knowledge bases for closed-book QA? In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1:… find
  10. Yanai Elazar, Nora Kassner, Shauli Ravfogel, Abhilasha Ravichander, Eduard Hovy, Hinrich Schütze, and Yoav Goldberg. 2021. Measuring and improving consistency in pretrained language models. Preprint, arXiv:2102.01017. find
  11. Shuzheng Gao, Xinjie Wen, Cuiyun Gao, Wenxuan Wang, and Michael R. Lyu. 2023. Constructing effective in-context demonstration for code intelligence tasks: An empirical study. ArXiv, abs/2304.07575. — [arXiv:2304.07575](https://arxiv.org/abs/2304.07575) find
  12. Cindy Gordon. 2023. Chatgpt is the fastest growing app in the history of web applications. https://www.forbes.com/sites/cindygordon/2023/02/02. Accessed: 2023-07-01. ↗
  13. Yu Gu, Sue Kase, Michelle Vanni, Brian Sadler, Percy Liang, Xifeng Yan, and Yu Su. 2021. Beyond i.i.d.: Three levels of generalization for question answering on knowledge bases. In Proceedings of the Web Conference 2021, WWW ’21, page 3477–3488. ACM. find
  14. Wenxiang Jiao, Wenxuan Wang, Jen-tse Huang, Xing Wang, Shuming Shi, and Zhaopeng Tu. 2023. Is chatgpt a good translator? yes with gpt-4 as the engine. arXiv preprint arXiv:2301.08745.
  15. Mario Karlovčec, Matjaž Krnc, and Riste Škrekovski. 2022. Evaluating group degree centrality and centralization in networks. Informatica, 46. find
  16. Nora Kassner, Philipp Dufter, and Hinrich Schütze. 2021a. Multilingual LAMA: Investigating knowledge in multilingual pretrained language models. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume, pages 3250–3258, Online. Assoc…
  17. Nora Kassner, Philipp Dufter, and Hinrich Schütze. 2021b. Multilingual lama: Investigating knowledge in multilingual pretrained language models. EACL.
  18. Md Tahmid Rahman Laskar, M Saiful Bari, Mizanur Rahman, Md Amran Hossen Bhuiyan, Shafiq R. Joty, and J. Huang. 2023. A systematic study and comprehensive evaluation of chatgpt on benchmark datasets. In Annual Meeting of the Association for Computational Linguistics.
  19. Patrick Lewis, Yuxiang Wu, Linqing Liu, Pasquale Minervini, Heinrich Kuttler, Aleksandra Piktus, Pontus Stenetorp, and Sebastian Riedel. 2021. Paq: 65 million probably-asked questions and what you can do with them. Transactions of the Association for Computational Linguistics, 9:1098–1115.
  20. Wangyue Li, Liangzhi Li, Tong Xiang, Xiao Liu, Wei Deng, and Noa Garcia. 2024. Can multiple-choice questions really be useful in detecting the abilities of llms? ArXiv, abs/2403.17752.
  21. Stephanie C. Lin, Jacob Hilton, and Owain Evans. 2021. Truthfulqa: Measuring how models mimic human falsehoods. In Annual Meeting of the Association for Computational Linguistics.
  22. Zixi Liu, Yang Feng, Yining Yin, J. Sun, Zhenyu Chen, and Baowen Xu. 2022. Qatest: A uniform fuzzing framework for question answering systems. Proceedings of the 37th IEEE/ACM International Conference on Automated Software Engineering. find
  23. Alex Mallen, Akari Asai, Victor Zhong, Rajarshi Das, Daniel Khashabi, and Hannaneh Hajishirzi. 2023. When not to trust language models: Investigating effectiveness of parametric and non-parametric memories. Preprint, arXiv:2212.10511. find
  24. R. Thomas McCoy, Ellie Pavlick, and Tal Linzen. 2019. Right for the wrong reasons: Diagnosing syntactic heuristics in natural language inference. ACL.
  25. Belinda Mo, Kyssen Yu, Joshua Kazdan, Joan Cabezas, Proud Mpala, Lisa Yu, Chris Cundy, Charilaos Kanatsoulis, and Sanmi Koyejo. 2025. Kggen: Extracting knowledge graphs from plain text with language models. arXiv preprint arXiv:2502.09956. — [arXiv:2502.09956](https://arxiv.org/abs/2502.09956) find
  26. Seyed Mahed Mousavi, Simone Alghisi, and Giuseppe Riccardi. 2024. Dyknow: Dynamically verifying time-sensitive factual knowledge in llms. Preprint, arXiv:2404.08700. and future directions towards knowledge graph chatbots. ArXiv, abs/2302.06466. find
  27. OpenAI. 2023. Gpt-4 technical report. abs/2303.08774.
  28. ArXiv, Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. Bleu: a method for automatic evaluation of machine translation. In Annual Meeting of the Association for Computational Linguistics.
  29. Fabio Petroni, Tim Rocktäschel, Patrick Lewis, Anton Bakhtin, Yuxiang Wu, Alexander H. Miller, and Sebastian Riedel. 2019. Language models as knowledge bases? ArXiv, abs/1909.01066.
  30. Daw Khin Po. 2020. Similarity based information retrieval using levenshtein distance algorithm. International Journal of Advances in Scientific Research and Engineering. find
  31. Qingchao Shen, Junjie Chen, J Zhang, Haoyu Wang, Shuang Liu, and Menghan Tian. 2022. Natural test generation for precise testing of question answering software. Proceedings of the 37th IEEE/ACM International Conference on Automated Software Engineering. find
  32. Kai Sun, Y. Xu, Hanwen Zha, Yue Liu, and Xinhsuai Dong. 2023. Head-to-tail: How knowledgeable are large language models (llms)? a.k.a. will llms replace knowledge graphs? ArXiv, abs/2308.10168. find
  33. Alon Talmor, Jonathan Herzig, Nicholas Lourie, and Jonathan Berant. 2019. Commonsenseqa: A question answering challenge targeting commonsense knowledge. NAACL.
  34. Yuqiao Tan, Shizhu He, Huanxuan Liao, Jun Zhao, and Kang Liu. 2025. Dynamic parametric retrieval augmented generation for test-time knowledge enhancement. Preprint, arXiv:2503.23895. — [arXiv:2503.23895](https://arxiv.org/abs/2503.23895) find
  35. Jason Wei, Nguyen Karina, Hyung Won Chung, Yunxin Joy Jiao, Spencer Papay, Amelia Glaese, John Schulman, and William Fedus. 2024. Measuring short-form factuality in large language models. Preprint, arXiv:2411.04368.
  36. Hao Wu, Wenxuan Wang, Yuxuan Wan, Wenxiang Jiao, and Michael R. Lyu. 2023. Chatgpt or grammarly? evaluating chatgpt on grammatical error correction benchmark. ArXiv, abs/2303.13648. find
  37. Shuai Zhang, Yi Tay, Lina Yao, and Qi Liu. 2019. Quaternion knowledge graph embeddings. Preprint, arXiv:1904.10281. find
  38. Chunting Zhou, Pengfei Liu, Puxin Xu, Srini Iyer, Jiao Sun, Yuning Mao, Xuezhe Ma, Avia Efrat, Ping Yu, Lili Yu, et al. 2023. Lima: Less is more for alignment. arXiv preprint arXiv:2305.11206. find
  39. Reham Omar, Omij Mangukiya, Panos Kalnis, and Essam Mansour. 2023. Chatgpt versus traditional question answering for knowledge graphs: Current status find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wang-2026-identifying-achilles-heel-iterative-method/