Elamur Research⌕
☰
home / papers / 2026 / Harnessing the Latent Space: From Steering Vectors to Model Calibrators for Control and Trust

Harnessing the Latent Space: From Steering Vectors to Model Calibrators for Control and Trust

Paper · 2026 · Nishant Subramani

Summary

This position-style overview addresses two linked problems: how to steer language models and how to know when their outputs deserve trust. Instead of treating models as black boxes judged only by behaviour, the author draws on four of his own studies that use hidden activations directly. For control, a single learned vector is added to a frozen model's internal states and optimised until the model reproduces a target sentence. This was first done with LSTM models trained on English Gigaword and later with small transformers, where concept-level steering was also explored. For trust, classifiers trained on internal features estimate whether a generation is correct. MICE does this for tool-calling with Llama models. ACUTE extends the idea to more model families and tasks and proposes a utility-oriented metric called EURO. Steering vectors recover sentences almost perfectly, and activation-based estimators beat standard post-hoc calibration, which tends to shrink toward base-rate predictions. Overall, the work argues that latent spaces can serve as practical levers for safer LLMs.

Key points

  • The LSTM language models were 2-layer networks (d = 256/512/1024) trained on 50M Gigaword sentences with a BPE vocabulary of 20,234 tokens. The large model reached token-level exact match ≥ 0.99, and even random token strings of up to 28 tokens were encoded nearly perfectly.
  • In transformers, injecting the steering vector into a middle layer (layer 6 self-attention or layer 7 feed-forward) at all timesteps gave BLEU-4 of 100. Injecting at the embedding layer gave only 33.99, and at the LM head only 6.72.
  • MICE was tested in a zero-shot setting by holding out each of 50 APIs in turn. It performed on par with post-hoc calibrators on smECE and ETCU while producing a wider spread of confidence values.
  • ACUTE was evaluated on six LLMs across MMLU, APIGen and SCITLDR. It achieved the best AUC-EURO while keeping smECE low. The strongest variants used late-layer activations or a PCA reduction to 20 components.
  • EURO ranges from an anti-oracle score of 0 to an oracle score of 1. It drops ETCU's assumption that abstaining always earns zero reward.

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

🎓Carnegie Mellon UniversityPittsburgh · United States598 works · 191,226 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

Cite

Cite as Nishant Subramani. 2026. Harnessing the Latent Space: From Steering Vectors to Model Calibrators for Control and Trust. Proceedings of The Big Picture v2: Crafting a Research Narrative.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.bigpicture-main.10

@inproceedings{subramani2026harnessing,
  title = {{Harnessing the Latent Space: From Steering Vectors to Model Calibrators for Control and Trust}},
  author = {Nishant Subramani},
  booktitle = {Proceedings of The Big Picture v2: Crafting a Research Narrative},
  year = {2026},
  doi = {10.18653/v1/2026.bigpicture-main.10},
  url = {https://doi.org/10.18653/v1/2026.bigpicture-main.10}
}

References

76 references, 28 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Dana Arad, Aaron Mueller, and Yonatan Belinkov. 2025. SAEs are good for steering – if you select the right features. In EMNLP.
  2. Neil Band, Xuechen Li, Tengyu Ma, and Tatsunori Hashimoto. 2024. Linguistic calibration of long-form generations. In ICML. find
  3. Richard E Barlow. 1972. Statistical inference under order restrictions: The theory and application of isotonic regression. (No Title). find
  4. Mohammad Beigi, Ying Shen, Runing Yang, Zihao Lin, Qifan Wang, Ankith Mohan, Jianfeng He, Ming Jin, Chang-Tien Lu, and Lifu Huang. 2024. InternalInspector i 2 : Robust confidence estimation in LLMs through internal states. In EMNLP Findings. find
  5. Nora Belrose, Zach Furman, Logan Smith, Danny Halawi, Igor V. Ostrovsky, Lev McKinney, Stella Biderman, and Jacob Steinhardt. 2023. Eliciting latent predictions from transformers with the tuned lens. arXiv.
  6. Eric Bigelow, Daniel Wurgaft, YingQiao Wang, Noah Goodman, Tomer Ullman, Hidenori Tanaka, and Ekdeep Singh Lubana. 2026. Belief dynamics reveal the dual nature of in-context learning and activation steering. Preprint, arXiv:2511.00617. — [arXiv:2511.00617](https://arxiv.org/abs/2511.00617) find
  7. Philip Gage. 1994. A new algorithm for data compression. The C Users Journal archive. find
  8. Yonatan Geifman, Guy Uziel, and Ran El-Yaniv. 2019. Bias-reduced uncertainty estimation for deep neural classifiers. In ICLR. find
  9. Mor Geva, Avi Caciularu, Kevin Wang, and Yoav Goldberg. 2022. Transformer feed-forward layers build predictions by promoting concepts in the vocabulary space. In EMNLP.
  10. Jarosław Błasiok and Preetum Nakkiran. 2024. Smooth ECE: Principled reliability diagrams via kernel smoothing. In ICLR. find
  11. W Brier Glenn. 1950. Verification of forecasts expressed in terms of probability. Monthly weather review, (1). find
  12. Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, and 1 others. 2020. Language models are few-shot learners. NeurIPS.
  13. David Graff and Christopher Cieri. 2003. English gigaword corpus. Linguistic Data Consortium. find
  14. Isabel Cachola, Kyle Lo, Arman Cohan, and Daniel Weld. 2020. TLDR: Extreme summarization of scientific documents. In EMNLP Findings.
  15. Daniel Cer, Mona Diab, Eneko Agirre, Iñigo Lopez-Gazpio, and Lucia Specia. 2017. SemEval-2017 task 1: Semantic textual similarity multilingual and crosslingual focused evaluation. In SemEval.
  16. Mauro Cettolo, Jan Niehues, Sebastian Stüker, Luisa Bentivogli, Rolando Cattoni, and Marcello Federico. 2016. find
  17. The IWSLT 2016 evaluation campaign. In Proceedings of the 13th International Conference on Spoken Language Translation. International Workshop on Spoken Language Translation. find
  18. Andrew M Dai and Quoc V Le. 2015. Semi-supervised sequence learning. NeurIPS. find
  19. A Philip Dawid. 1982. The well-calibrated bayesian. Journal of the American statistical Association, (379). find
  20. Shrey Desai and Greg Durrett. 2020. Calibration of pre-trained transformers. In EMNLP. find
  21. 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,…
  22. Jacob Dunefsky and Arman Cohan. 2025. One-shot optimized steering vectors mediate safety-relevant behaviors in llms. In COLM. find
  23. Mostafa Elhoushi, Akshat Shrivastava, Diana Liskovich, Basil Hosmer, Bram Wasti, Liangzhen Lai, Anas Mahmoud, Bilge Acun, Saurabh Agarwal, Ahmed Roman, Ahmed Aly, Beidi Chen, and Carole-Jean Wu. 2024. LayerSkip: Enabling early exit inference and self-speculative decoding. In ACL.
  24. Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, Amy Yang, Angela Fan, Anirudh Goyal, Anthony Hartshorn, Aobo Yang, Archi Mitra, Archie Sravankumar, Artem Korenev, Arthur Hinsvark, and 22 ot… — The Llama 3 Herd of Models
  25. Alex Graves. 2012. Sequence transduction with recurrent neural networks. arXiv.
  26. Dirk Groeneveld, Iz Beltagy, Evan Walsh, Akshita Bhagia, Rodney Kinney, Oyvind Tafjord, Ananya Jha, Hamish Ivison, Ian Magnusson, Yizhong Wang, Shane Arora, David Atkinson, Russell Authur, Khyathi Chandu, Arman Cohan, Jennifer Dumas, Yanai Elazar, Yuling Gu, Jack Hessel, and 22 others. 2024. OLMo: A… find
  27. Pengcheng He, Xiaodong Liu, Jianfeng Gao, and Weizhu Chen. 2021. Deberta: Decoding-enhanced bert with disentangled attention. In ICLR.
  28. Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. 2021. Measuring massive multitask language understanding. In ICLR.
  29. Sepp Hochreiter and Jürgen Schmidhuber. 1997. Long short-term memory. Neural computation, (8). find
  30. 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. TACL. find
  31. Diederik P Kingma and Jimmy Ba. 2015. Adam: A method for stochastic optimization. In ICLR.
  32. Meelis Kull, Telmo Silva Filho, and Peter Flach. 2017. Beta calibration: a well-founded and easily implemented improvement on logistic calibration for binary classifiers. In Proceedings of the 20th International Conference on Artificial Intelligence and Statistics, Proceedings of Machine Learning Re… find
  33. Kenneth Li, Oam Patel, Fernanda Viégas, Hanspeter Pfister, and Martin Wattenberg. 2023. Inference-time intervention: Eliciting truthful answers from a language model. NeurIPS.
  34. Percy Liang, Rishi Bommasani, Tony Lee, Dimitris Tsipras, Dilara Soylu, Michihiro Yasunaga, Yian Zhang, Deepak Narayanan, Yuhuai Wu, Ananya Kumar, Benjamin Newman, Binhang Yuan, Bobby Yan, Ce Zhang, Christian Cosgrove, Christopher D. Manning, Christopher R’e, Diana Acosta-Navas, Drew A. Hudson, and… find
  35. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. Bleu: a method for automatic evaluation of machine translation. In ACL.
  36. Matthew E. Peters, Mark Neumann, Mohit Iyyer, Matt Gardner, Christopher Clark, Kenton Lee, and Luke Zettlemoyer. 2018. Deep contextualized word representations. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Techno…
  37. John Platt. 1999. Probabilistic outputs for support vector machines and comparisons to regularized likelihood methods. Advances in large margin classifiers, (3). find
  38. Ofir Press and Lior Wolf. 2017. Using the output embedding to improve language models. In Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 2, Short Papers.
  39. Jiarui Liu, Jivitesh Jain, Mona Diab, and Nishant Subramani. 2025. Llm microscope: What model internals reveal about answer correctness and context utilization. arXiv. find
  40. Changle Qu, Sunhao Dai, Xiaochi Wei, Hengyi Cai, Shuaiqiang Wang, Dawei Yin, Jun Xu, and Jirong Wen. 2024. Tool learning with large language models: a survey. Frontiers of Computer Science.
  41. Zuxin Liu, Thai Hoang, Jianguo Zhang, Ming Zhu, Tian Lan, Juntao Tan, Weiran Yao, Zhiwei Liu, Yihao Feng, Rithesh RN, and 1 others. 2024. Apigen: Automated pipeline for generating verifiable and diverse function-calling datasets. NeurIPS. find
  42. Alec Radford, Jeff Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. 2019. Language models are unsupervised multitask learners. find
  43. J.I. Marcum. 1960. A statistical theory of target detection by pulsed radar. IRE Transactions on Information Theory. find
  44. Jack Merullo, Carsten Eickhoff, and Ellie Pavlick. 2024. Language models implement simple Word2Vec-style vector arithmetic. In NAACL. find
  45. George Morgulis and John Hewitt. 2026. Subliminal steering: Stronger encoding of hidden signals. arXiv. find
  46. Allan H Murphy and Edward S Epstein. 1967. Verification of probabilistic predictions: A brief review. Journal of Applied Meteorology and Climatology, (5). find
  47. Elizbar A Nadaraya. 1964. On estimating regression. Theory of Probability & Its Applications, (1). find
  48. Mahdi Pakdaman Naeini, Gregory Cooper, and Milos Hauskrecht. 2015. Obtaining well calibrated probabilities using bayesian binning. In AAAI, 1. find
  49. Alexandru Niculescu-Mizil and Rich Caruana. 2005. Predicting good probabilities with supervised learning. In ICML. find
  50. Andrew Nobel. 1996. Histogram regression estimation using data-dependent partitions. The Annals of Statistics, (3). nostalgebraist. 2020. Interpreting GPT: The logit lens. Blogpost. find
  51. Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J. Liu. 2020. Exploring the limits of transfer learning with a unified text-to-text transformer. JMLR.
  52. Tal Schuster, Adam Fisch, Jai Gupta, Mostafa Dehghani, Dara Bahri, Vinh Tran, Yi Tay, and Donald Metzler. 2022. Confident adaptive language modeling. NeurIPS. find
  53. Rico Sennrich, Barry Haddow, and Alexandra Birch. 2016. Neural machine translation of rare words with subword units. In ACL.
  54. Tianxiao Shen, Tao Lei, Regina Barzilay, and T. Jaakkola. 2017. Style transfer from non-parallel text by cross-alignment. In NIPS.
  55. Aarohi Srivastava, Abhinav Rastogi, Abhishek Rao, Abu Awal Md Shoeb, Abubakar Abid, Adam Fisch, Adam R. Brown, Adam Santoro, Aditya Gupta, Adrià Garriga-Alonso, Agnieszka Kluska, Aitor Lewkowycz, Akshat Agarwal, Alethea Power, Alex Ray, Alex Warstadt, Alexander W. Kocurek, Ali Safaya, Ali Tazarv, an… — Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models
  56. Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. 2014. Dropout: A simple way to prevent neural networks from overfitting. JMLR, (56).
  57. Elias Stengel-Eskin and Benjamin Van Durme. 2023. Calibrated interpretation: Confidence estimation in semantic parsing. TACL. find
  58. Nishant Subramani, Samuel Bowman, and Kyunghyun Cho. 2019. Can unconditional language models recover arbitrary sentences? NeurIPS. find
  59. Nishant Subramani, Jason Eisner, Justin Svegliato, Benjamin Van Durme, Yu Su, and Sam Thomson. 2025a. MICE for CATs: Model-internal confidence estimation for calibrating agents with tools. In NAACL. find
  60. Nishant Subramani, Alfredo Gomez, and Mona T. Diab. 2025b. SimBA: Simplifying benchmark analysis using performance matrices alone. In EMNLP Findings. find
  61. Nishant Subramani, Palash Goyal, Yiwen Song, Mani Malek, Yuan Xue, Tomas Pfister, and Hamid Palangi. 2026. The acute protocol: Operationalizing language model activations for better calibration, utility, and trust. In ICML. find
  62. Nishant Subramani, Nivedita Suresh, and Matthew Peters. 2022. Extracting latent steering vectors from pretrained language models. In ACL Findings.
  63. Gemma Team, Aishwarya Kamath, Johan Ferret, Shreya Pathak, Nino Vieillard, Ramona Merhej, Sarah Perrin, Tatiana Matejovicova, Alexandre Ramé, Morgane Rivière, and 1 others. 2025. Gemma 3 technical report. arXiv. find
  64. Stephen Wright and Jorge Nocedal. 1999. Numerical optimization. Springer Science, (67-68). find
  65. Daniel Wurgaft, Can Rager, Matthew Kowal, Vasudev Shyam, Sheridan Feucht, Usha Bhalla, Tal Haklay, Eric Bigelow, Raphael Sarfati, Thomas McGrath, Owen Lewis, Jack Merullo, Noah Goodman, Thomas Fel, Atticus Geiger, and Ekdeep Singh Lubana. 2026. Manifold steering reveals the shared geometry of neural… — [arXiv:2605.05115](https://arxiv.org/abs/2605.05115) find
  66. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, and 1 others. 2025. Qwen3 technical report. arXiv.
  67. Alexander Yom Din, Taelin Karidi, Leshem Choshen, and Mor Geva. 2024. Jump to conclusions: Shortcutting transformers with linear transformations. In COLING. ELRA and ICCL. find
  68. Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q Weinberger, and Yoav Artzi. 2020. Bertscore: Evaluating text generation with bert. In ICLR. find
  69. Ruiqi Zhong, Charlie Snell, Dan Klein, and Jason Eisner. 2023. Non-programmers can label programs indirectly via active examples: A case study with text-to-SQL. In EMNLP. find
  70. Ian Tenney, Dipanjan Das, and Ellie Pavlick. 2019. BERT rediscovers the classical NLP pipeline. In ACL.
  71. Arun James Thirunavukarasu, Darren S. J. Ting, Kabilan Elangovan, Laura Gutierrez, Ting Fang Tan, and Daniel Shu Wei Ting. 2023. Large language models in medicine. Nature Medicine. find
  72. Alexander Matt Turner, Lisa Thiergart, Gavin Leech, David Udell, Juan J Vazquez, Ulisse Mini, and Monte MacDiarmid. 2023. Steering language models with activation engineering. arXiv.
  73. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In NeurIPS.
  74. John G. Wagner and James W. Ayres. 1977. Bioavailability assessment: Methods to estimate total area (auc 0-∞) and total amount excreted (ae∞) and importance of blood and urine sampling scheme with application to digoxin. Journal of Pharmacokinetics and Biopharmaceutics. find
  75. Boshi Wang, Hao Fang, Jason Eisner, Benjamin Van Durme, and Yu Su. 2024. LLMs in the imaginarium: Tool learning through simulated trial and error. In ACL. find
  76. Geoffrey S Watson. 1964. Smooth regression analysis. Sankhyā: The Indian Journal of Statistics, Series A. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/subramani-2026-harnessing-latent-space-steering-vectors/