Elamur Research⌕
☰
home / papers / 2025 / Influences on LLM Calibration: A Study of Response Agreement, Loss Functions, and Prompt Styles

Influences on LLM Calibration: A Study of Response Agreement, Loss Functions, and Prompt Styles

Paper · 2025 · Yuxi Xia, Pedro Henrique Luz De Araujo, Klim Zaporojets et al.

Summary

LLMs often express confidence that does not match how often they are actually right. Earlier calibration methods were usually tested on only a few models and prompt formats. This paper runs a controlled comparison across 12 LLMs from five model families (2B–72B parameters), four prompt styles (Verbalized, Zero-shot, Chain-of-Thought, Few-shot) and four open-ended question answering datasets, including NQ.

The authors propose Calib-n, an auxiliary confidence estimator built on bert-base-uncased. It reads the answers of n target LLMs to the same question together, so it can use agreement or disagreement between models as an uncertainty signal. They train it with binary cross-entropy, focal loss or an AUC surrogate loss. Answer correctness is labelled by an LLM judge.

No single method wins in every setting. Counted over all settings, however, auxiliary models beat raw token probabilities, verbalized confidence and the APRICOT baseline. Focal loss and response agreement both help, and few-shot prompts work best. This gives practical guidance for producing reliable confidence scores from LLMs.

Key points

  • Models: seven small (2–9B) and five large (27–72B) LLMs. Correctness labels come from the judge model Prometheus-8x7b-v2.0, which outputs a binary correct/incorrect score.
  • Calib-n architecture: a BERT backbone with a classification head mapping n·768 to n outputs, followed by a sigmoid. Each input is a question[SEP]answer pair, one per target LLM, and the model returns one confidence per answer.
  • Loss functions: (FL)Calib-1, a single target LLM trained with focal loss, achieved the best overall results. The authors recommend Calib-n when calibrating several LLMs, since one auxiliary model serves them all and it reduces overconfidence.
  • Example result: on NQ with large models, (BCE)Calib-n reached an ECE of 0.049 on Llama3.1-70b. Platt scaling lowered ECE-t but compressed confidences into a narrow range.
  • Robustness: auxiliary models stayed well calibrated as accuracy varied, whereas internal probabilities and verbalized confidence tended to be overconfident even on datasets with accuracy below 40%. Code and data are available on GitHub (Influences-on-LLM-Calibration).

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 ViennaVienna · Austria18 works · 38,729 citations🎓Aarhus UniversityAarhus · Denmark7 works · 12 citations

 Fields, topics and concepts

📚 Shelves 4The 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 works

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

Cite

Cite as Yuxi Xia, Pedro Henrique Luz De Araujo, Klim Zaporojets and Benjamin Roth. 2025. Influences on LLM Calibration: A Study of Response Agreement, Loss Functions, and Prompt Styles. 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.188

@inproceedings{xia2025influences,
  title = {{Influences on LLM Calibration: A Study of Response Agreement, Loss Functions, and Prompt Styles}},
  author = {Yuxi Xia, Pedro Henrique Luz {De Araujo}, Klim Zaporojets and Benjamin Roth},
  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.188},
  url = {https://doi.org/10.18653/v1/2025.acl-long.188}
}

References

36 references, 14 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Marah Abdin et al. 2024. Phi-3 technical report: A highly capable language model locally on your phone. ArXiv preprint, abs/2404.14219.
  2. Glenn W. Brier. 1950. Verification of forecasts expressed in terms of probability. Monthly Weather Review, 78(1):1 – 3. find
  3. Banghao Chen, Zhaofeng Zhang, Nicolas Langrené, and Shengxin Zhu. 2023. Unleashing the potential of prompt engineering in large language models: a comprehensive review. find
  4. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies,…
  5. 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…
  6. Aaron Grattafiori et al. 2024. The llama 3 herd of models. ArXiv preprint, abs/2407.21783.
  7. Chuan Guo, Geoff Pleiss, Yu Sun, and Kilian Q. Weinberger. 2017a. On calibration of modern neural networks. In Proceedings of the 34th International Conference on Machine Learning, ICML 2017, Sydney, NSW, Australia, 6-11 August 2017, volume 70 of Proceedings of Machine Learning Research, pages 1321–… find
  8. Chuan Guo, Geoff Pleiss, Yu Sun, and Kilian Q. Weinberger. 2017b. On calibration of modern neural networks. In Proceedings of the 34th International Conference on Machine Learning, ICML 2017, Sydney, NSW, Australia, 6-11 August 2017, volume 70 of Proceedings of Machine Learning Research, pages 1321–… find
  9. Albert Q. Jiang, Alexandre Sablayrolles, Antoine Roux, Arthur Mensch, Blanche Savary, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Emma Bou Hanna, Florian Bressand, Gianna Lengyel, Guillaume Bour, Guillaume Lample, Lélio Renard Lavaud, Lucile Saulnier, Marie-Anne Lachaux, Pierre Stock,… find
  10. Zhengbao Jiang, Jun Araki, Haibo Ding, and Graham Neubig. 2021. How can we know when language models know? on the calibration of language models for question answering. Transactions of the Association for Computational Linguistics, 9:962–977. find
  11. 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…
  12. Jaeyoung Kim, Dongbin Na, Sungchul Choi, and Sung-bin Lim. 2023. Bag of tricks for in-distribution calibration of pretrained transformers. In Findings of the Association for Computational Linguistics: EACL 2023, pages 551–563, Dubrovnik, Croatia. Association for Computational Linguistics. find
  13. Seungone Kim, Juyoung Suk, Shayne Longpre, Bill Yuchen Lin, Jamin Shin, Sean Welleck, Graham Neubig, Moontae Lee, Kyungjae Lee, and Minjoon Seo. 2024. Prometheus 2: An open source language model specialized in evaluating other language models.
  14. Tom Kwiatkowski, Jennimaria Palomaki, Olivia Redfield, Michael Collins, Ankur Parikh, Chris Alberti, Danielle Epstein, Illia Polosukhin, Jacob Devlin, Kenton Lee, Kristina Toutanova, Llion Jones, Matthew Kelcey, Ming-Wei Chang, Andrew M. Dai, Jakob Uszkoreit, Quoc Le, and Slav Petrov. 2019. Natural…
  15. Stephanie Lin, Jacob Hilton, and Owain Evans. 2022. Teaching models to express their uncertainty in words. Transactions on Machine Learning Research. find
  16. Tsung-Yi Lin, Priya Goyal, Ross B. Girshick, Kaiming He, and Piotr Dollár. 2017. Focal loss for dense object detection. In IEEE International Conference on Computer Vision, ICCV 2017, Venice, Italy, October 22-29, 2017, pages 2999–3007. IEEE Computer Society.
  17. Xin Liu, Muhammad Khalifa, and Lu Wang. 2024. Litcab: Lightweight language model calibration over short- and long-form responses. In The Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024. Open-Review.net. find
  18. Sewon Min, Xinxi Lyu, Ari Holtzman, Mikel Artetxe, Mike Lewis, Hannaneh Hajishirzi, and Luke Zettlemoyer. 2022. Rethinking the role of demonstrations: What makes in-context learning work? In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 11048–11064, Ab…
  19. Jooyoung Moon, Jihyo Kim, Younghak Shin, and Sangheum Hwang. 2020. Confidence-aware learning for deep neural networks. In Proceedings of the 37th International Conference on Machine Learning, ICML 2020, 13-18 July 2020, Virtual Event, volume 119 of Proceedings of Machine Learning Research, pages 703… find
  20. Jishnu Mukhoti, Viveka Kulharia, Amartya Sanyal, Stuart Golodetz, Philip H. S. Torr, and Puneet K. Dokania. 2020. Calibrating deep neural networks using focal loss. In Advances in Neural Information Processing Systems 33: Annual Conference on Neural Information Processing Systems 2020, NeurIPS 2020,… find
  21. Shiyu Ni, Keping Bi, Lulu Yu, and Jiafeng Guo. 2024. Are large language models more honest in their probabilistic or verbalized confidence? find
  22. John Platt. 1999. Probabilistic outputs for support vector machines and comparisons to regularized likelihood methods. Advances in Large Margin Classifiers, 10(3):61–74. find
  23. Inioluwa Deborah Raji, Andrew Smart, Rebecca N White, Margaret Mitchell, Timnit Gebru, Ben Hutchinson, Jamila Smith-Loud, Daniel Theron, and Parker Barnes. 2020. Closing the ai accountability gap: Defining an end-to-end framework for internal algorithmic auditing. In Proceedings of the 2020 conferen… find
  24. Gemma Team et al. 2024. Gemma 2: Improving open language models at a practical size. ArXiv preprint, abs/2408.00118. find
  25. Katherine Tian, Eric Mitchell, Allan Zhou, Archit Sharma, Rafael Rafailov, Huaxiu Yao, Chelsea Finn, and Christopher Manning. 2023. Just ask for calibration: Strategies for eliciting calibrated confidence scores from language models fine-tuned with human feedback. In Proceedings of the 2023 Conferen… find
  26. Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Canton Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes, Jeremy Fu, Wenyin Fu, Brian Fuller,… — Llama 2: Open Foundation and Fine-Tuned Chat Models
  27. Dennis Ulmer, Martin Gubri, Hwaran Lee, Sangdoo Yun, and Seong Oh. 2024. Calibrating large language models using their generations only. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 15440–15459, Bangkok, Thailand. Associati… find
  28. Lean Wang, Lei Li, Damai Dai, Deli Chen, Hao Zhou, Fandong Meng, Jie Zhou, and Xu Sun. 2023. Label words are anchors: An information flow perspective for understanding in-context learning. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 9840–9855, Sin…
  29. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed H. Chi, Quoc V. Le, and Denny Zhou. 2022. Chain-of-thought prompting elicits reasoning in large language models. In Advances in Neural Information Processing Systems 35: Annual Conference on Neural Information Processi…
  30. Yi Yang, Wen-tau Yih, and Christopher Meek. 2015. WikiQA: A challenge dataset for open-domain question answering. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 2013–2018, Lisbon, Portugal. Association for Computational Linguistics.
  31. Zhuoning Yuan, Yan Yan, Milan Sonka, and Tianbao Yang. 2021. Large-scale robust deep AUC maximization: A new surrogate loss and empirical studies on medical image classification. In 2021 IEEE/CVF International Conference on Computer Vision, ICCV 2021, Montreal, QC, Canada, October 10-17, 2021, pages… find
  32. Min Zhang, Jianfeng He, Taoran Ji, and Chang-Tien Lu. 2024. Don’t go to extremes: Revealing the excessive sensitivity and calibration limitations of LLMs in implicit hate speech detection. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Pape… find
  33. Johannes Welbl, Nelson F. Liu, and Matt Gardner. 2017. Crowdsourcing multiple choice science questions. In Proceedings of the 3rd Workshop on Noisy User-generated Text, pages 94–106, Copenhagen, Denmark. Association for Computational Linguistics.
  34. Yuxi Xia, Kilm Zaporojets, and Benjamin Roth. 2024. Black-box model ensembling for textual and visual question answering via information fusion. find
  35. Miao Xiong, Zhiyuan Hu, Xinyang Lu, Yifei Li, Jie Fu, Junxian He, and Bryan Hooi. 2024. Can llms express their uncertainty? an empirical evaluation of confidence elicitation in llms. In The Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024. Open… find
  36. An Yang, Baosong Yang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Zhou, Chengpeng Li, Chengyuan Li, Dayiheng Liu, Fei Huang, Guanting Dong, Haoran Wei, Huan Lin, Jialong Tang, Jialin Wang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Ma, Jianxin Yang, Jin Xu, Jingren Zhou, Jinze Bai, Jinzheng He, Juny… find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/xia-2025-influences-llm-calibration-study-response/