Elamur Research⌕
☰
home / papers / 2025 / Reviewing Scientific Papers for Critical Problems With Reasoning LLMs: Baseline Approaches and Automatic Evaluation

Reviewing Scientific Papers for Critical Problems With Reasoning LLMs: Baseline Approaches and Automatic Evaluation

Paper · 2025 · Tianmai M. Zhang and Neil F. Abernethy

Summary

Peer review is strained by rising submission volumes, and having LLMs write entire reviews invites misuse and manipulation. Zhang and Abernethy instead treat LLMs as manuscript quality checkers with one narrow job: flagging critical errors or unsound reasoning that could undermine a paper's conclusions. They build WithdrarXiv-Check, a benchmark of arXiv papers that their authors withdrew while naming the specific mistake. The papers were filtered with Gemini 2.5 Flash and then by manual review.

Two baselines give the paper to a reasoning model, either as an attached PDF or as LaTeX source inside the prompt, and ask for the five most serious problems. Expert annotation is costly, so evaluation is automated. Two strong LLM judges check whether any reported problem matches the authors' stated error (hit rate) and estimate how many reported problems are genuine (precision). Among models available in mid-2025, o3 found the most critical errors at moderate API cost. The work contributes a reusable dataset, code, and evaluation recipe for document-level scientific reasoning, while insisting that humans remain central to peer review.

Key points

  • Dataset construction: The source is WithdrarXiv's "critical errors" category (6,018 candidates). An LLM screen reduced this to 2,190 papers, and manual exclusions left 1,225 papers. Of these, 245 form the test set and 980 are reserved for training and validation.
  • Subject mix and formats: The data is dominated by math (about 50%) and physics, with page counts centred around 14 pages. LaTeX source exists for 88% of test papers.
  • Models compared: Gemini 2.5 Pro, Gemini 2.5 Flash, o3, o4-mini and Claude 3.7 Sonnet were tested without tool or web access, using up to k = 5 problems per paper.
  • Judging setup: Gemini 2.5 Pro and o3 served as judges, and a hit or true positive required both to agree. The metrics are HR@k and AP@k, or MHR@k and MAP@k when runs are repeated. Claude 3.7 Sonnet was dropped as a judge because its PDF hit rate was low.
  • Findings and release: o3 performed best overall, and the o-series models consistently used all five problem slots. Stated limitations include closed-source models only, single runs, no human calibration of the judges, and a math/physics-heavy domain. The dataset, code and model outputs are released on GitHub.

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 WashingtonSeattle · United States402 works · 212,762 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksHistory and Philosophy of Science · 62 works

🧭 Topics 2Research subjects and languages the work is about.Language models & pretraining · 3,894 worksReasoning · 1,685 works

Cite

Cite as Tianmai M. Zhang and Neil F. Abernethy. 2025. Reviewing Scientific Papers for Critical Problems With Reasoning LLMs: Baseline Approaches and Automatic Evaluation. arXiv:2505.23824.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2505.23824 · arXiv 2505.23824

@misc{zhang2025reviewing,
  title = {{Reviewing Scientific Papers for Critical Problems With Reasoning LLMs: Baseline Approaches and Automatic Evaluation}},
  author = {Tianmai M. Zhang and Neil F. Abernethy},
  year = {2025},
  doi = {10.48550/arxiv.2505.23824},
  eprint = {2505.23824},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2505.23824}
}

References

29 references, 8 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Howard Bauchner and Frederick P. Rivara. Use of artificial intelligence and the future of peer review. Health Affairs Scholar, 2(5), 2024. doi: 10.1093/haschl/qxae058. — [doi:10.1093/haschl/qxae058](https://doi.org/10.1093/haschl/qxae058) find
  2. Mike D’Arcy, Tom Hope, Larry Birnbaum, and Doug Downey. MARG: Multi-agent review generation for scientific papers, 2024. URL https://arxiv.org/abs/2401.04259. ↗
  3. Jiangshu Du, Yibo Wang, Wenting Zhao, Zhongfen Deng, Shuaiqi Liu, Renze Lou, Henry Peng Zou, Pranav Narayanan Venkit, Nan Zhang, Mukund Srinath, Haoran Ranran Zhang, Vipul Gupta, Yinghui Li, Tao Li, Fei Wang, Qin Liu, Tianlin Liu, Pengzhi Gao, Congying Xia, Chen Xing, Cheng Jiayang, Zhaowei Wang, Yi… — LLMs Assist NLP Researchers: Critique Paper (Meta-)Reviewing
  4. Michael Duan, Anshuman Suri, Niloofar Mireshghallah, Sewon Min, Weijia Shi, Luke Zettlemoyer, Yulia Tsvetkov, Yejin Choi, David Evans, and Hannaneh Hajishirzi. Do membership inference attacks work on large language models? In Conference on Language Modeling (COLM), 2024. URL https://arxiv.org/abs/24… ↗
  5. Yujuan Fu, Ozlem Uzuner, Meliha Yetisgen, and Fei Xia. Does data contamination detection work (well) for LLMs? a survey and evaluation on detection assumptions. In Luis Chiruzzo, Alan Ritter, and Lu Wang, editors, Findings of the Association for Computational Linguistics: NAACL 2025, pages 5235–5256… find
  6. Zhaolin Gao, Kianté Brantley, and Thorsten Joachims. Reviewer2: Optimizing review generation through prompt generation, 2024. URL https://arxiv.org/abs/2402.10886. ↗
  7. Jaeho Kim, Yunseok Lee, and Seulki Lee. Position: The AI conference peer review crisis demands author feedback and reviewer rewards. In Proceedings of the 42nd International Conference on Machine Learning, Vancouver, Canada, 2025. PMLR. URL https://arxiv.org/abs/2505. 04966. ↗
  8. Ilia Kuznetsov, Osama Mohammed Afzal, Koen Dercksen, Nils Dycke, Alexander Goldberg, Tom Hope, Dirk Hovy, Jonathan K. Kummerfeld, Anne Lauscher, Kevin Leyton-Brown, Sheng Lu, Mausam, Margot Mieskes, Aurélie Névéol, Danish Pruthi, Lizhen Qu, Roy Schwartz, Noah A. Smith, Thamar Solorio, Jingyan Wang,… find
  9. Diana Kwon. Is it ok for AI to write science papers? Nature survey shows researchers are split. Nature, 641, 2025. doi: 10.1038/d41586-025-01463-8. find
  10. Weixin Liang, Yuhui Zhang, Hancheng Cao, Binglu Wang, Daisy Yi Ding, Xinyu Yang, Kailas Vodrahalli, Siyu He, Daniel Scott Smith, Yian Yin, Daniel A. McFarland, and James Zou. Can large language models provide useful feedback on research papers? a large-scale empirical analysis. NEJM AI, 1(8), 2024.…
  11. Jialiang Lin, Jiaxin Song, Zhangping Zhou, Yidong Chen, and Xiaodong Shi. Automated scholarly paper review: Concepts, technologies, and challenges. Information Fusion, 98(101830), 2023. doi: 10.1016/j.inffus.2023.101830. — [doi:10.1016/j.inffus.2023.101830](https://doi.org/10.1016/j.inffus.2023.101830) find
  12. Zhicheng Lin. Hidden prompts in manuscripts exploit AI-assisted peer review, 2025. URL https://arxiv.org/abs/2507.06185. ↗
  13. Ryan Liu and Nihar B. Shah. ReviewerGPT? an exploratory study on using large language models for paper reviewing, 2023. URL https://arxiv.org/abs/2306.00622. ↗
  14. Miryam Naddaf. AI is transforming peer review — and many scientists are worried. Nature, 639, 2025. doi: 10.1038/d41586-025-00894-7. find
  15. Michèle B. Nuijten and Joshua R. Polanin. "statcheck": automatically detect statistical reporting inconsistencies to increase reproducibility of meta-analyses. Research Synthesis Methods, 11(5), 2020. doi: 10.1002/jrsm.1408. — [doi:10.1002/jrsm.1408](https://doi.org/10.1002/jrsm.1408) find
  16. Delip Rao, Jonathan Young, Thomas Dietterich, and Chris Callison-Burch. WithdrarXiv: A large-scale dataset for retraction study, 2024. URL https://arxiv.org/abs/2412.03775. — [arXiv:2412.03775](https://arxiv.org/abs/2412.03775) ↗
  17. Maria Janina Sarol, Shufan Ming, Shruthan Radhakrishna, Jodi Schneider, and Halil Kilicoglu. Assessing citation integrity in biomedical publications: corpus annotation and NLP models. Bioinformatics, 40(7), 2024. doi: 10.1093/bioinformatics/btae420. — [doi:10.1093/bioinformatics/btae420](https://doi.org/10.1093/bioinformatics/btae420) find
  18. Hyungyu Shin, Jingyu Tang, Yoonjoo Lee, Nayoung Kim, Hyunseung Lim, Ji Yong Cho, Hwajung Hong, Moontae Lee, and Juho Kim. Automatically evaluating the paper reviewing capability of large language models, 2025. URL https://arxiv.org/abs/2502.17086. — [arXiv:2502.17086](https://arxiv.org/abs/2502.17086) ↗
  19. Guijin Son, Jiwoo Hong, Honglu Fan, Heejeong Nam, Hyunwoo Ko, Seungwon Lim, Jinyeop Song, Jinha Choi, Gonçalo Paulo, Youngjae Yu, and Stella Biderman. When AI co-scientists fail: SPOT-a benchmark for automated verification of scientific research, 2025. URL https://arxiv.org/abs/2505.11855. ↗
  20. Pawin Taechoyotin, Guanchao Wang, Tong Zeng, Bradley Sides, and Daniel Acuna. MAMORX: Multi-agent multi-modal scientific review generation with external knowledge. In Neurips 2024 Workshop Foundation Models for Science: Progress, Opportunities, and Challenges, Vancouver, Canada, 2024. URL https://op… find
  21. Cheng Tan, Dongxin Lyu, Siyuan Li, Zhangyang Gao, Jingxuan Wei, Siqi Ma, Zicheng Liu, and Stan Z. Li. Peer review as a multi-turn and long-context dialogue with role-based interactions, 2024. URL https://arxiv.org/abs/2406.05688. — [arXiv:2406.05688](https://arxiv.org/abs/2406.05688) ↗
  22. Keith Tyser, Ben Segev, Gaston Longhitano, Xin-Yu Zhang, Zachary Meeks, Jason Lee, Uday Garg, Nicholas Belsten, Avi Shporer, Madeleine Udell, Dov Te’eni, and Iddo Drori. AI-driven review systems: Evaluating LLMs in scalable and bias-aware academic reviews, 2024. URL https://arxiv.org/abs/2408.10365. ↗
  23. Rui Ye, Xianghe Pang, Jingyi Chai, Jiaao Chen, Zhenfei Yin, Zhen Xiang, Xiaowen Dong, Jing Shao, and Siheng Chen. Are we there yet? revealing the risks of utilizing large language models in scholarly peer review, 2024. URL https://arxiv.org/abs/2412.01708. ↗
  24. Jianxiang Yu, Zichen Ding, Jiaqi Tan, Kangyang Luo, Zhenmin Weng, Chenghua Gong, Long Zeng, RenJing Cui, Chengcheng Han, Qiushi Sun, Zhiyong Wu, Yunshi Lan, and Xiang Li. Automated peer reviewing in paper SEA: Standardization, evaluation, and analysis. In Yaser Al-Onaizan, Mohit Bansal, and Yun-Nung… find
  25. Sihang Zeng, Kai Tian, Kaiyan Zhang, Yuru wang, Junqi Gao, Runze Liu, Sa Yang, Jingxuan Li, Xinwei Long, Jiaheng Ma, Biqing Qi, and Bowen Zhou. ReviewRL: Towards automated scientific review with RL, 2025. URL https://arxiv.org/abs/2508.10308. ↗
  26. Tianmai M. Zhang and Neil F. Abernethy. Detecting reference errors in scientific literature with large language models, 2024. URL https://arxiv.org/abs/2411.06101. — [arXiv:2411.06101](https://arxiv.org/abs/2411.06101) ↗
  27. Ruiyang Zhou, Lu Chen, and Kai Yu. Is LLM a reliable reviewer? a comprehensive evaluation of LLM on automatic paper reviewing tasks. In Nicoletta Calzolari, Min-Yen Kan, Veronique Hoste, Alessandro Lenci, Sakriani Sakti, and Nianwen Xue, editors, Proceedings of the 2024 Joint International Conferenc…
  28. Minjun Zhu, Yixuan Weng, Linyi Yang, and Yue Zhang. Deepreview: Improving llm-based paper review with human-like deep thinking process, 2025. URL https://arxiv.org/abs/2503. 08569. ↗
  29. Zhenzhen Zhuang, Jiandong Chen, Hongfeng Xu, Yuwen Jiang, and Jialiang Lin. Large language models for automated scholarly paper review: A survey. Information Fusion, 124(103332), 2025. doi: 10.1016/j.inffus.2025.103332. — [doi:10.1016/j.inffus.2025.103332](https://doi.org/10.1016/j.inffus.2025.103332) find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zhang-2025-reviewing-scientific-papers-critical-problems/