Elamur Research⌕
☰
home / papers / 2025 / Confidence v.s. Critique: A Decomposition of Self-Correction Capability for LLMs

Confidence v.s. Critique: A Decomposition of Self-Correction Capability for LLMs

Paper · 2025 · Zhe Yang, Yichang Zhang, Yudong Wang et al.

Summary

Intrinsic self-correction means asking an LLM to re-check and revise its own answer without external feedback. It sometimes improves accuracy and sometimes lowers it, and prior work disagrees about its value. Yang et al. cross answer correctness before and after revision, which gives four outcomes: confident, unconfident, critical and stubborn. They reduce these to two abilities: confidence (keeping right answers) and critique (repairing wrong ones).

They formalize these as Confidence Level (CL) and Critique Score (CS). Each is a conditional probability of a correct revised answer, given a correct or a wrong first answer. They show that post-correction accuracy is a weighted combination of the two, and they add a normalized overall measure, the Relative Self-correction Score (RSS).

Experiments on many LLMs reveal "conservative" and "liberal" model profiles. Prompting or in-context learning tends to raise one ability only by lowering the other. Their training recipe, Confidence and Critique improvement Tuning (CCT), restructures SFT data so the model learns the final answer from a question plus an initial answer. CCT lifts both CL and CS and clearly beats vanilla SFT on post-correction accuracy.

Key points

  • Metrics: CL = P(correct after | correct before) and CS = P(correct after | wrong before), estimated by repeated sampling. Acc 2 (accuracy after correction) decomposes as a weighted sum of CL and CS. RSS places Acc 2 between derived lower and upper bounds; RSS = 0.5 means correction leaves accuracy unchanged, and values above 0.5 mean a gain.
  • Findings across models:
    • Self-correction usually, but not always, helps.
    • Confidence is typically stronger than critique.
    • Models from the same family behave alike, possibly because of shared pre-training data.
  • Prompting and ICL reveal a trade-off: neither raises CL and CS together without fine-tuning.
  • CCT mixes confidence-oriented (CLT) and critique-oriented (CST) data; neither works alone. On BoolQ, a share of CLT data of about 40% gave the best self-correction. Combining SFT with CCT achieves both high Acc 1 and high Acc 2.
  • Code: github.com/Zhe-Young/SelfCorrectDecompose. Stated limitation: the metrics need costly repeated sampling and cannot separate stubbornness from questions that are simply too hard for the model.

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

🎓Peking UniversityBeijing · China244 works · 39,359 citations🏢Alibaba GroupHangzhou · China168 works · 15,386 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

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

Cite

Cite as Zhe Yang, Yichang Zhang, Yudong Wang, Ziyao Xu, Junyang Lin and Zhifang Sui. 2025. Confidence v.s. Critique: A Decomposition of Self-Correction Capability for LLMs. 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.203

@inproceedings{yang2025confidence,
  title = {{Confidence v.s. Critique: A Decomposition of Self-Correction Capability for LLMs}},
  author = {Zhe Yang, Yichang Zhang, Yudong Wang, Ziyao Xu, Junyang Lin and Zhifang Sui},
  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.203},
  url = {https://doi.org/10.18653/v1/2025.acl-long.203}
}

References

51 references, 27 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. 2023. Gpt-4 technical report. arXiv preprint arXiv:2303.08774.
  2. AI@Meta. 2024. Llama 3 model card. find
  3. Jinze Bai, Shuai Bai, Yunfei Chu, Zeyu Cui, Kai Dang, Xiaodong Deng, Yang Fan, Wenbin Ge, Yu Han, Fei Huang, et al. 2023. Qwen technical report. arXiv preprint arXiv:2309.16609.
  4. Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. 2020. Language models are few-shot learners. Advances in neural information processing systems, 33:1877–1901.
  5. Yupeng Chang, Xu Wang, Jindong Wang, Yuan Wu, Linyi Yang, Kaijie Zhu, Hao Chen, Xiaoyuan Yi, Cunxiang Wang, Yidong Wang, Wei Ye, Yue Zhang, Yi Chang, Philip S. Yu, Qiang Yang, and Xing Xie. 2023. A survey on evaluation of large language models. Preprint, arXiv:2307.03109.
  6. Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde de Oliveira Pinto, Jared Kaplan, Harri Edwards, Yuri Burda, Nicholas Joseph, Greg Brockman, Alex Ray, Raul Puri, Gretchen Krueger, Michael Petrov, Heidy Khlaaf, Girish Sastry, Pamela Mishkin, Brooke Chan, Scott Gray, Nick Ryder, Mikhai… — Evaluating Large Language Models Trained on Code
  7. Xinyun Chen, Maxwell Lin, Nathanael Schärli, and Denny Zhou. 2024. Teaching large language models to self-debug. In The Twelfth International Conference on Learning Representations. 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 Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics:…
  9. Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, Christopher Hesse, and John Schulman. 2021. Training verifiers to solve math word problems. arXiv preprint arXiv:2110.14168.
  10. DeepSeek-AI. 2024. Deepseek llm: Scaling open-source language models with longtermism. arXiv preprint arXiv:2401.02954. find
  11. Qingxiu Dong, Lei Li, Damai Dai, Ce Zheng, Jingyuan Ma, Rui Li, Heming Xia, Jingjing Xu, Zhiyong Wu, Baobao Chang, Xu Sun, Lei Li, and Zhifang Sui. 2024. A survey on in-context learning. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 1107–1128, Miami…
  12. Bofei Gao, Feifan Song, Yibo Miao, Zefan Cai, Zhe Yang, Liang Chen, Helan Hu, Runxin Xu, Qingxiu Dong, Ce Zheng, Shanghaoran Quan, Wen Xiao, Ge Zhang, Daoguang Zan, Keming Lu, Bowen Yu, Dayiheng Liu, Zeyu Cui, Jian Yang, Lei Sha, Houfeng Wang, Zhifang Sui, Peiyi Wang, Tianyu Liu, and Baobao Chang. 2… — [arXiv:2409.02795](https://arxiv.org/abs/2409.02795) find
  13. Bofei Gao, Feifan Song, Zhe Yang, Zefan Cai, Yibo Miao, Qingxiu Dong, Lei Li, Chenghao Ma, Liang Chen, Runxin Xu, et al. 2024b. Omni-math: A universal olympiad level mathematic benchmark for large language models. arXiv preprint arXiv:2410.07985.
  14. Jiahui Geng, Fengyu Cai, Yuxia Wang, Heinz Koeppl, Preslav Nakov, and Iryna Gurevych. 2024. A survey of confidence estimation and calibration in large language models. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language… find
  15. Team GLM, Aohan Zeng, Bin Xu, Bowen Wang, Chenhui Zhang, Da Yin, Dan Zhang, Diego Rojas, Guanyu Feng, Hanlin Zhao, et al. 2024. Chatglm: A family of large language models from glm-130b to glm-4 all tools. arXiv preprint arXiv:2406.12793.
  16. Haixia Han, Jiaqing Liang, Jie Shi, Qianyu He, and Yanghua Xiao. 2024. Small language model can self-correct. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 18162– 18170. find
  17. Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. 2021. Measuring massive multitask language understanding. In International Conference on Learning Representations.
  18. Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2021. Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685.
  19. Jie Huang, Xinyun Chen, Swaroop Mishra, Huaixiu Steven Zheng, Adams Wei Yu, Xinying Song, and Denny Zhou. 2024. Large language models cannot self-correct reasoning yet. In The Twelfth International Conference on Learning Representations.
  20. Hamish Ivison, Yizhong Wang, Jiacheng Liu, Zeqiu Wu, Valentina Pyatkin, Nathan Lambert, Noah A. Smith, Yejin Choi, and Hannaneh Hajishirzi. 2024. Unpacking dpo and ppo: Disentangling best practices for learning from preference feedback. Preprint, arXiv:2406.09279.
  21. Albert Q Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, et al. 2023a. Mistral 7b. arXiv preprint arXiv:2310.06825. find
  22. Dongwei Jiang, Jingyu Zhang, Orion Weller, Nathaniel Weir, Benjamin Van Durme, and Daniel Khashabi. 2024. Self-[in] correct: Llms struggle with refining self-generated responses. CoRR. find
  23. Shuyang Jiang, Yuhao Wang, and Yu Wang. 2023b. Selfevolve: A code evolution framework via large language models. arXiv preprint arXiv:2306.02907. — [arXiv:2306.02907](https://arxiv.org/abs/2306.02907) find
  24. Ryo Kamoi, Yusen Zhang, Nan Zhang, Jiawei Han, and Rui Zhang. 2024. When can LLMs actually correct their own mistakes? a critical survey of self-correction of LLMs. Transactions of the Association for Computational Linguistics, 12:1417–1440.
  25. Aviral Kumar, Vincent Zhuang, Rishabh Agarwal, Yi Su, John D Co-Reyes, Avi Singh, Kate Baumli, Shariq Iqbal, Colton Bishop, Rebecca Roelofs, et al. 2024. Training language models to self-correct via reinforcement learning. arXiv preprint arXiv:2409.12917.
  26. Loka Li, Zhenhao Chen, Guangyi Chen, Yixuan Zhang, Yusheng Su, Eric Xing, and Kun Zhang. 2024. Confidence matters: Revisiting intrinsic self-correction capabilities of large language models. arXiv preprint arXiv:2402.12563. — [arXiv:2402.12563](https://arxiv.org/abs/2402.12563) find
  27. Dancheng Liu, Amir Nassereldine, Ziming Yang, Chenhui Xu, Yuting Hu, Jiajie Li, Utkarsh Kumar, Changjae Lee, and Jinjun Xiong. 2024. Large language models have intrinsic self-correction ability. CoRR. find
  28. Aman Madaan, Niket Tandon, Prakhar Gupta, Skyler Hallinan, Luyu Gao, Sarah Wiegreffe, Uri Alon, Nouha Dziri, Shrimai Prabhumoye, Yiming Yang, et al. 2024. Self-refine: Iterative refinement with self-feedback. Advances in Neural Information Processing Systems, 36. find
  29. Long Ouyang, Jeff Wu, Xu Jiang, Diogo Almeida, Carroll L. Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul Christiano, Jan Leike, and Ryan Lowe. 2022. Train…
  30. Keyu Pan and Yawen Zeng. 2023. Do llms possess a personality? making the mbti test an amazing evaluation for large language models. Preprint, arXiv:2307.16180.
  31. Liangming Pan, Michael Saxon, Wenda Xu, Deepak Nathani, Xinyi Wang, and William Yang Wang. 2024. Automatically correcting large language models: Surveying the landscape of diverse automated correction strategies. Transactions of the Association for Computational Linguistics, 12:484–506.
  32. Yuxiao Qu, Tianjun Zhang, Naman Garg, and Aviral Kumar. 2024. Recursive introspection: Teaching foundation model agents how to self-improve. In Automated Reinforcement Learning: Exploring Meta-Learning, AutoML, and LLMs. find
  33. Alec Radford, Karthik Narasimhan, Tim Salimans, Ilya Sutskever, et al. 2018. Improving language understanding by generative pre-training. find
  34. Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, Ilya Sutskever, et al. 2019. Language models are unsupervised multitask learners. OpenAI blog, 1(8):9. find
  35. Rafael Rafailov, Archit Sharma, Eric Mitchell, Stefano Ermon, Christopher D. Manning, and Chelsea Finn. 2024. Direct preference optimization: Your language model is secretly a reward model. Preprint, arXiv:2305.18290.
  36. Kaya Stechly, Matthew Marquez, and Subbarao Kambhampati. 2023. Gpt-4 doesn’t know it’s wrong: An analysis of iterative prompting for reasoning problems. In NeurIPS 2023 Foundation Models for Decision Making Workshop. find
  37. Alon Talmor, Jonathan Herzig, Nicholas Lourie, and Jonathan Berant. 2019. CommonsenseQA: A question answering challenge targeting commonsense knowledge. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies,…
  38. Gladys Tyen, Hassan Mansoor, Victor Carbune, Peter Chen, and Tony Mak. 2024. LLMs cannot find reasoning errors, but can correct them given the error location. In Findings of the Association for Computational Linguistics: ACL 2024, pages 13894–13908, Bangkok, Thailand. Association for Computational L…
  39. Karthik Valmeekam, Matthew Marquez, and Subbarao Kambhampati. 2023. Can large language models really improve by self-critiquing their own plans? In NeurIPS 2023 Foundation Models for Decision Making Workshop. find
  40. Jason Wei, Maarten Bosma, Vincent Zhao, Kelvin Guu, Adams Wei Yu, Brian Lester, Nan Du, Andrew M Dai, and Quoc V Le. 2021. Finetuned language models are zero-shot learners. In International Conference on Learning Representations.
  41. Sean Welleck, Ximing Lu, Peter West, Faeze Brahman, Tianxiao Shen, Daniel Khashabi, and Yejin Choi. 2023. Generating sequences by learning to self-correct. In The Eleventh International Conference on Learning Representations.
  42. Zhenyu Wu, Qingkai Zeng, Zhihan Zhang, Zhaoxuan Tan, Chao Shen, and Meng Jiang. 2024. Large language models can self-correct with key condition verification. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 12846–12867, Miami, Florida, USA. Association… find
  43. Zhiheng Xi, Senjie Jin, Yuhao Zhou, Rui Zheng, Songyang Gao, Jia Liu, Tao Gui, Qi Zhang, and Xuanjing Huang. 2023. Self-Polish: Enhance reasoning in large language models via problem refinement. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 11383–11406, Singapore. A… find
  44. Yuchen Yan, Jin Jiang, Yang Liu, Yixin Cao, Xin Xu, Xunliang Cai, Jian Shao, et al. 2024. Sc-math: Spontaneous step-level self-correction makes large language models better mathematical reasoners. arXiv preprint arXiv:2409.01524. — [arXiv:2409.01524](https://arxiv.org/abs/2409.01524) find
  45. An Yang, Baosong Yang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Zhou, Chengpeng Li, Chengyuan Li, Dayiheng Liu, Fei Huang, et al. 2024a. Qwen2 technical report. arXiv preprint arXiv:2407.10671. find
  46. Zhe Yang, Damai Dai, Peiyi Wang, and Zhifang Sui. 2023. Not all demonstration examples are equally beneficial: Reweighting demonstration examples for in-context learning. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 13209–13221, Singapore. Association for Computati… find
  47. Zhe Yang, Yichang Zhang, Tianyu Liu, Jian Yang, Junyang Lin, Chang Zhou, and Zhifang Sui. 2024b. Can large language models always solve easy problems if they can solve harder ones? In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 1531–1555, Miami, Flor… find
  48. Shengyu Zhang, Linfeng Dong, Xiaoya Li, Sen Zhang, Xiaofei Sun, Shuhe Wang, Jiwei Li, Runyi Hu, Tianwei Zhang, Fei Wu, et al. 2023. Instruction tuning for large language models: A survey. arXiv preprint arXiv:2308.10792.
  49. Wenqi Zhang, Yongliang Shen, Linjuan Wu, Qiuying Peng, Jun Wang, Yueting Zhuang, and Weiming Lu. 2024a. Self-contrast: Better reflection through inconsistent solving perspectives. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), page…
  50. Yunxiang Zhang, Muhammad Khalifa, Lajanugen Logeswaran, Jaekyeom Kim, Moontae Lee, Honglak Lee, and Lu Wang. 2024b. Small language models need strong verifiers to self-correct reasoning. In Findings of the Association for Computational Linguistics: ACL 2024, pages 15637–15653, Bangkok, Thailand. Ass… find
  51. Jeffrey Zhou, Tianjian Lu, Swaroop Mishra, Siddhartha Brahma, Sujoy Basu, Yi Luan, Denny Zhou, and Le Hou. 2023. Instruction-following evaluation for large language models. arXiv preprint arXiv:2311.07911.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/yang-2025-confidence-v-s-critique-decomposition/