Elamur Research⌕
☰
home / papers / 2025 / The Mirage of Model Editing: Revisiting Evaluation in the Wild

The Mirage of Model Editing: Revisiting Evaluation in the Wild

Paper · 2025 · Wanli Yang, Fei Sun, Jiajun Tan et al.

Summary

Published model-editing results often look nearly perfect. This paper asks whether those results hold once the evaluation reflects how an edited LLM would actually be used. The authors build QAEdit, a benchmark drawn from Natural Questions, TriviaQA and SimpleQA. GPT-4 adds the subjects and paraphrased prompts that editing methods need, and ZsRE supplies the locality pairs. They also propose WILD (Without Intervention, Live Decoding), a task-agnostic protocol that follows standard QA evaluation in lm-evaluation-harness and lets the edited model generate freely.

Under WILD, six editing methods on three LLMs perform far worse than the literature suggests. A controlled analysis breaks the evaluation pipeline into four components: what the model receives as input, how text is decoded, where outputs are cut off, and how answers are scored. It shows that earlier inflated scores come mainly from teacher forcing and from cutting outputs to the length of the gold answer, since both reveal the ground truth to the model. Sequential and batched editing break down within about 1000 edits.

The work matters because it shows that the field's evaluation habits overstate progress, and it supplies a stricter benchmark and protocol for future editing research.

Key points

  • Single editing: on Llama-2-7b-chat, average success under WILD is 38.5%, compared with 96.8% under synthetic evaluation.
  • Benchmark and setup: QAEdit contains 19,249 samples in ten categories. The tested methods are FT-M, MEND, ROME, MEMIT, GRACE and WISE, evaluated alongside the ZsRE and CounterFact benchmarks.
  • Four flaws in synthetic evaluation:
    • the same prompt is used for editing and testing;
    • teacher forcing feeds the gold answer during decoding;
    • outputs are truncated to the target length;
    • a match-ratio metric gives credit for partially correct answers.
  • Sequential editing:
    • after 1000 edits, average success falls to about 10%;
    • average locality is only 21.3%;
    • GRACE preserves unrelated knowledge but rarely injects the new facts;
    • FT-M keeps relatively stable reliability.
  • Mini-batch editing:
    • no batch size pushes any method above 30% at 1000 edits;
    • MEMIT works best when all edits go in one batch, while FT-M works best with a batch size of 1;
    • later edits erode knowledge injected by earlier batches.

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

🎓University of Chinese Academy of SciencesBeijing · China147 works · 8,480 citations🔬Institute of Computing TechnologyBeijing · China51 works · 3,411 citations

 Fields, topics and concepts

📚 Shelves 4The 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 works

Cite

Cite as Wanli Yang, Fei Sun, Jiajun Tan, Xinyu Ma, Qi Cao, Dawei Yin, Huawei Shen and Xueqi Cheng. 2025. The Mirage of Model Editing: Revisiting Evaluation in the Wild. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2025.acl-long.745

@inproceedings{yang2025mirage,
  title = {{The Mirage of Model Editing: Revisiting Evaluation in the Wild}},
  author = {Wanli Yang, Fei Sun, Jiajun Tan, Xinyu Ma, Qi Cao, Dawei Yin, Huawei Shen and Xueqi Cheng},
  booktitle = {Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2025},
  doi = {10.18653/v1/2025.acl-long.745},
  url = {https://doi.org/10.18653/v1/2025.acl-long.745}
}

References

48 references, 11 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Canyu Chen, Baixiang Huang, Zekun Li, Zhaorun Chen, Shiyang Lai, Xiongxiao Xu, Jia-Chen Gu, Jindong Gu, Huaxiu Yao, Chaowei Xiao, Xifeng Yan, William Yang Wang, Philip Torr, Dawn Song, and Kai Shu. 2024a. Can editing llms inject harm? Preprint, arXiv:2407.20224. — [arXiv:2407.20224](https://arxiv.org/abs/2407.20224) find
  2. Ruizhe Chen, Yichen Li, Zikai Xiao, and Zuozhu Liu. 2024b. Large language model bias mitigation from the perspective of knowledge editing. In ICLR 2024 Workshop on Secure and Trustworthy Large Language Models. find
  3. Nicola De Cao, Wilker Aziz, and Ivan Titov. 2021. Editing factual knowledge in language models. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 6491– 6506, Online and Punta Cana, Dominican Republic. Association for Computational Linguistics. find
  4. Qingxiu Dong, Damai Dai, Yifan Song, Jingjing Xu, Zhifang Sui, and Lei Li. 2022. Calibrating factual knowledge in pretrained language models. In Findings of the Association for Computational Linguistics: EMNLP 2022, pages 5937–5947, Abu Dhabi, United Arab Emirates. Association for Computational Ling…
  5. Junfeng Fang, Houcheng Jiang, Kun Wang, Yunshan Ma, Jie Shi, Xiang Wang, Xiangnan He, and Tat-Seng Chua. 2025. Alphaedit: Null-space constrained model editing for language models. In The Thirteenth International Conference on Learning Representations. find
  6. Leo Gao, Jonathan Tow, Baber Abbasi, Stella Biderman, Sid Black, Anthony DiPofi, Charles Foster, Laurence Golding, Jeffrey Hsu, Alain Le Noac’h, Haonan Li, Kyle McDonell, Niklas Muennighoff, Chris Ociepa, Jason Phang, Laria Reynolds, Hailey Schoelkopf, Aviya Skowron, Lintang Sutawika, Eric Tang, Ani… find
  7. Mor Geva, Avi Caciularu, Kevin Wang, and Yoav Goldberg. 2022. Transformer feed-forward layers build predictions by promoting concepts in the vocabulary space. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 30–45, Abu Dhabi, United Arab Emirates. Asso…
  8. Mor Geva, Roei Schuster, Jonathan Berant, and Omer Levy. 2021. Transformer feed-forward layers are key-value memories. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 5484–5495, Online and Punta Cana, Dominican Republic. Association for Computational… find
  9. Eleventh International Conference on Learning Representations. find
  10. Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, and et al. 2024. The llama 3 herd of models. Preprint, arXiv:2407.21783.
  11. Albert Q. Jiang, Alexandre Sablayrolles, Arthur Mensch, and et al. 2023. Mistral 7b. Preprint, arXiv:2310.06825. find
  12. Jia-Chen Gu, Hao-Xiang Xu, Jun-Yu Ma, Pan Lu, Zhen-Hua Ling, Kai-Wei Chang, and Nanyun Peng. 2024. Model editing harms general abilities of large language models: Regularization to the rescue. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 16801– 168…
  13. Mandar Joshi, Eunsol Choi, Daniel Weld, and Luke Zettlemoyer. 2017. TriviaQA: A large scale distantly supervised challenge dataset for reading comprehension. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1601–1611, Vancouver…
  14. Akshat Gupta, Sidharth Baskaran, and Gopala Anumanchipalli. 2024a. Rebuilding ROME : Resolving model collapse during sequential model editing. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 21738–21744, Miami, Florida, USA. Association for Computatio… find
  15. Akshat Gupta, Anurag Rao, and Gopala Anumanchipalli. 2024b. Model editing at scale leads to gradual and catastrophic forgetting. In Findings of the Association for Computational Linguistics: ACL 2024, pages 15202–15232, Bangkok, Thailand. Association for Computational Linguistics. find
  16. Akshat Gupta, Dev Sajnani, and Gopala Anumanchipalli. 2024c. A unified framework for model editing. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 15403– 15418, Miami, Florida, USA. Association for Computational Linguistics. find
  17. Zeyu Han, Chao Gao, Jinyang Liu, Jeff Zhang, and Sai Qian Zhang. 2024. Parameter-efficient fine-tuning for large models: A comprehensive survey. Transactions on Machine Learning Research.
  18. Thomas Hartvigsen, Swami Sankaranarayanan, Hamid Palangi, Yoon Kim, and Marzyeh Ghassemi. 2023. Aging with GRACE: Lifelong model editing with discrete key-value adaptors. In Thirty-seventh Conference on Neural Information Processing Systems. find
  19. Jason Hoelscher-Obermaier, Julia Persson, Esben Kran, Ioannis Konstas, and Fazl Barez. 2023. Detecting edit failures in large language models: An improved specificity benchmark. In Findings of the Association for Computational Linguistics: ACL 2023, pages 11548–11559, Toronto, Canada. Association fo… find
  20. Baixiang Huang, Canyu Chen, Xiongxiao Xu, Ali Payani, and Kai Shu. 2025. Can knowledge editing really correct hallucinations? In The Thirteenth International Conference on Learning Representations. find
  21. Zeyu Huang, Yikang Shen, Xiaofeng Zhang, Jie Zhou, Wenge Rong, and Zhang Xiong. 2023. Transformer-patcher: One mistake worth one neuron. In The Tom Kwiatkowski, Jennimaria Palomaki, Olivia Redfield, and et al. 2019. Natural questions: A benchmark for question answering research. Transactions of the…
  22. Omer Levy, Minjoon Seo, Eunsol Choi, and Luke Zettlemoyer. 2017. Zero-shot relation extraction via reading comprehension. In Proceedings of the 21st Conference on Computational Natural Language Learning (CoNLL 2017), pages 333–342, Vancouver, Canada. Association for Computational Linguistics.
  23. Dawei Li, Bohan Jiang, Liangjie Huang, Alimohammad Beigi, Chengshuai Zhao, Zhen Tan, Amrita Bhattacharjee, Yuxuan Jiang, Canyu Chen, Tianhao Wu, et al. 2024a. From generation to judgment: Opportunities and challenges of llm-as-a-judge. arXiv preprint arXiv:2411.16594. find
  24. Xiaopeng Li, Shasha Li, Shezheng Song, Jing Yang, Jun Ma, and Jie Yu. 2024b. Pmet: precise model editing in a transformer. In Proceedings of the Thirty-Eighth AAAI Conference on Artificial Intelligence and Thirty-Sixth Conference on Innovative Applications of Artificial Intelligence and Fourteenth S… find
  25. Zhoubo Li, Ningyu Zhang, Yunzhi Yao, Mengru Wang, Xi Chen, and Huajun Chen. 2024c. Unveiling the pitfalls of knowledge editing for large language models. In The Twelfth International Conference on Learning Representations. find
  26. Jun-Yu Ma, Hong Wang, Hao-Xiang Xu, Zhen-Hua Ling, and Jia-Chen Gu. 2025. Perturbation-restrained sequential model editing. In The Thirteenth International Conference on Learning Representations. find
  27. Kevin Meng, David Bau, Alex J Andonian, and Yonatan Belinkov. 2022. Locating and editing factual associations in GPT. In Advances in Neural Information Processing Systems. find
  28. Kevin Meng, Arnab Sen Sharma, Alex J Andonian, Yonatan Belinkov, and David Bau. 2023. Mass-editing memory in a transformer. In The Eleventh International Conference on Learning Representations. find
  29. Meta. 2024. Introducing meta llama 3: The most capable openly available llm to date. https://ai.meta.com/blog/meta-llama-3/. ↗
  30. Eric Mitchell, Charles Lin, Antoine Bosselut, Chelsea Finn, and Christopher D Manning. 2022. Fast model editing at scale. In International Conference on Learning Representations. find
  31. Tsung-Hsuan Pan, Chung-Chi Chen, Hen-Hsen Huang, and Hsin-Hsi Chen. 2024. “why” has the least side effect on model editing. Preprint, arXiv:2409.18679. — [arXiv:2409.18679](https://arxiv.org/abs/2409.18679) find
  32. Chenmien Tan, Ge Zhang, and Jie Fu. 2024. Massive editing for large language models via meta learning. In International Conference on Learning Representations. find
  33. Hugo Touvron, Louis Martin, Kevin Stone, and et al. 2023. Llama 2: Open foundation and fine-tuned chat models. Preprint, arXiv:2307.09288.
  34. Haoyu Wang, Tianci Liu, Ruirui Li, Monica Xiao Cheng, Tuo Zhao, and Jing Gao. 2024a. RoseLoRA: Row and column-wise sparse low-rank adaptation of pre-trained language model for knowledge editing and fine-tuning. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing… find
  35. Peng Wang, Zexi Li, Ningyu Zhang, Ziwen Xu, Yunzhi Yao, Yong Jiang, Pengjun Xie, Fei Huang, and Huajun Chen. 2024b. WISE: Rethinking the knowledge memory for lifelong model editing of large language models. In The Thirty-eighth Annual Conference on Neural Information Processing Systems. find
  36. Peng Wang, Ningyu Zhang, Bozhong Tian, and et al. 2024c. EasyEdit: An easy-to-use knowledge editing framework for large language models. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations), pages 82–93, Bangkok, Thailand. Assoc… find
  37. Song Wang, Yaochen Zhu, Haochen Liu, Zaiyi Zheng, Chen Chen, and Jundong Li. 2024d. Knowledge editing for large language models: A survey. ACM Computing Surveys, 57(3):1–37.
  38. 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.
  39. Xiaobao Wu, Liangming Pan, William Yang Wang, and Anh Tuan Luu. 2024. AKEW: Assessing knowledge editing in the wild. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 15118–15133, Miami, Florida, USA. Association for Computational Linguistics. find
  40. Xinwei Wu, Junzhuo Li, Minghui Xu, Weilong Dong, Shuangzhi Wu, Chao Bian, and Deyi Xiong. 2023. DEPN: Detecting and editing privacy neurons in pre-trained language models. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 2875–2886, Singapore. Associati… find
  41. Wanli Yang, Fei Sun, Xinyu Ma, Xun Liu, Dawei Yin, and Xueqi Cheng. 2024a. The butterfly effect of model editing: Few edits can trigger large language models collapse. In Findings of the Association for Computational Linguistics: ACL 2024, pages 5419– 5437, Bangkok, Thailand. Association for Computa… find
  42. Wanli Yang, Fei Sun, Jiajun Tan, Xinyu Ma, Du Su, Dawei Yin, and Huawei Shen. 2024b. The fall of ROME: Understanding the collapse of LLMs in model editing. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 4079–4087, Miami, Florida, USA. Association for Computational Li… find
  43. Yunzhi Yao, Peng Wang, Bozhong Tian, Siyuan Cheng, Zhoubo Li, Shumin Deng, Huajun Chen, and Ningyu Zhang. 2023. Editing large language models: Problems, methods, and opportunities. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 10222–10240, Singapore… find
  44. Lang Yu, Qin Chen, Jie Zhou, and Liang He. 2024. Melo: enhancing model editing with neuron-indexed dynamic lora. In Proceedings of the Thirty-Eighth AAAI Conference on Artificial Intelligence and Thirty-Sixth Conference on Innovative Applications of Artificial Intelligence and Fourteenth Symposium o… find
  45. Ningyu Zhang, Yunzhi Yao, Bozhong Tian, and et al. 2024. A comprehensive study of knowledge editing for large language models. Preprint, arXiv:2401.01286. find
  46. Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q Weinberger, and Yoav Artzi. 2020. Bertscore: Evaluating text generation with bert. In International Conference on Learning Representations. find
  47. Ce Zheng, Lei Li, Qingxiu Dong, Yuxuan Fan, Zhiyong Wu, Jingjing Xu, and Baobao Chang. 2023. Can we edit factual knowledge by in-context learning? In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 4862–4876, Singapore. Association for Computational Ling… find
  48. Chen Zhu, Ankit Singh Rawat, Manzil Zaheer, Srinadh Bhojanapalli, Daliang Li, Felix Yu, and Sanjiv Kumar. 2020. Modifying memories in transformer models. Preprint, arXiv:2012.00363. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/yang-2025-mirage-model-editing-revisiting-evaluation/