Elamur Research⌕
☰
home / papers / 2026 / How Controllable Are Large Language Models? A Unified Evaluation across Behavioral Granularities

How Controllable Are Large Language Models? A Unified Evaluation across Behavioral Granularities

Paper · 2026 · Ziwen Xu, Kewei Xu, Haoming Xu et al.

Summary

LLMs used in sensitive settings can drift in intent, tone, or persona. Existing steering benchmarks cover isolated behaviors, define concepts inconsistently, and ignore how specific a control target is. The authors propose SteerEval, a benchmark that crosses three behavioral domains (language features, sentiment, personality) with three specification levels adapted from Marr's levels of analysis:

  • L1 fixes the intended direction.
  • L2 fixes the strategy for conveying it.
  • L3 requires a concrete, checkable surface marker such as a specific word or token.

An automated LLM pipeline produces the data. It generates hierarchical concepts, rewrites questions so they do not reveal the target, and builds paired preference samples. All data then pass automatic filtering and human review. Prompt-based and activation-based steering methods, run through EasyEdit, are scored by an LLM judge on concept expression, instruction following, and fluency. Steering works fairly reliably at L1 and forces a trade-off at L2. At L3 it mostly fails, even when general output quality is sacrificed. The benchmark therefore shows where current steering techniques stop working.

Key points

  • SteerEval is organized as 3 domains × 3 granularity levels (L1 computational, L2 algorithmic, L3 implementational). For example, increasing redundancy in language features ends at L3 with a requirement to include "(i.e.,".
  • Unlike AXBENCH, whose concepts come from SAE feature descriptions and whose prompts are sampled from Alpaca-Eval, SteerEval uses explicit behavioral definitions. It also provides concept-targeted preference pairs and concept-aligned evaluation sets.
  • The steering methods compared include prompt-based steering and activation-based steering via concept vectors. All are implemented in the open-source EasyEdit framework, and the code is released in the zjunlp/EasyEdit repository.
  • The authors acknowledge several limitations:
    • The setup is restricted to single-turn, single-concept control.
    • It does not cover multi-turn, tool-use, or safety-critical domains.
    • Hyperparameter tuning is not claimed to be optimal.
    • Scoring depends on an LLM judge.
  • The paper was published in ACL 2026 (Volume 1: Long Papers), pp. 31269–31299, by authors from Zhejiang University and Alibaba Group.

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

🏢Alibaba GroupHangzhou · China201 works · 17,173 citations🎓Zhejiang UniversityHangzhou · China224 works · 13,587 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 2Research subjects and languages the work is about.Language models & pretraining · 3,894 worksEvaluation & benchmarks · 1,341 works

Cite

Cite as Ziwen Xu, Kewei Xu, Haoming Xu, Haiwen Hong, Longtao Huang, Hui Xue, Ningyu Zhang, Yongliang Shen, Guozhou Zheng, Huajun Chen and Shumin Deng. 2026. How Controllable Are Large Language Models? A Unified Evaluation across Behavioral Granularities. Proceedings of the 64th 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/v1/2026.acl-long.1443

@inproceedings{xu2026how,
  title = {{How Controllable Are Large Language Models? A Unified Evaluation across Behavioral Granularities}},
  author = {Ziwen Xu, Kewei Xu, Haoming Xu, Haiwen Hong, Longtao Huang, Hui Xue, Ningyu Zhang, Yongliang Shen, Guozhou Zheng, Huajun Chen and Shumin Deng},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-long.1443},
  url = {https://doi.org/10.18653/v1/2026.acl-long.1443}
}

References

66 references, 20 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Usman Anwar, Abulhair Saparov, Javier Rando, Daniel Paleka, Miles Turpin, Peter Hase, Ekdeep Singh Lubana, Erik Jenner, Stephen Casper, Oliver Sourbut, Benjamin L. Edelman, Zhaowei Zhang, Mario Günther, Anton Korinek, José Hernández-Orallo, Lewis Hammond, Eric J. Bigelow, Alexander Pan, Lauro Langos… — Foundational Challenges in Assuring Alignment and Safety of Large Language Models
  2. Andy Arditi, Owain Evans, and Jack Lindsey. 2025. Persona vectors: Monitoring and controlling character traits in language models. http://arxiv.org/abs/2507.21509. — [arXiv:2507.21509](https://arxiv.org/abs/2507.21509) ↗
  3. Amos Azaria, Rina Azoulay, and Shulamit Reches. 2024. Chatgpt is a remarkable tool-for experts. DATA INTELLIGENCE, 6(1):240–296. find
  4. David Badre. 2025. Cognitive control. Annual Review of Psychology, 76(Volume 76, 2025):167–195. find
  5. Lukasz Bartoszcze, Sarthak Munshi, Bryan Sukidi, Jennifer Yen, Zejia Yang, David Williams-King, Linh Le, Kosi Asuzu, and Carsten Maple. 2025. Representation engineering for large-language models: Survey and research challenges. CoRR, abs/2502.17601. — [arXiv:2502.17601](https://arxiv.org/abs/2502.17601) find
  6. Reza Bayat, Ali Rahimi-Kalahroudi, Mohammad Pezeshki, Sarath Chandar, and Pascal Vincent. 2025. Steering large language model activations in sparse spaces. CoRR, abs/2503.00177. — [arXiv:2503.00177](https://arxiv.org/abs/2503.00177) find
  7. Daniel Beaglehole, Adityanarayanan Radhakrishnan, Enric Boix-Adsera, and Mikhail Belkin. 2026. Toward universal steering and monitoring of ai models. Science, 391(6787):787–792. find
  8. Eric J. Bigelow, Daniel Wurgaft, YingQiao Wang, Noah D. Goodman, Tomer D. Ullman, Hidenori Tanaka, and Ekdeep Singh Lubana. 2025. Belief dynamics reveal the dual nature of in-context learning and activation steering. CoRR, abs/2511.00617. — [arXiv:2511.00617](https://arxiv.org/abs/2511.00617) find
  9. Matthew Botvinick and Todd Braver. 2015. Motivation and cognitive control: From behavior to neural mechanism. Annual Review of Psychology, 66(Volume 66, 2015):83–113. find
  10. Matthew M Botvinick. 2008. Hierarchical models of behavior and prefrontal function. Trends in cognitive sciences, 12(5):201–208. find
  11. Trenton Bricken, Adly Templeton, Joshua Batson, Brian Chen, Adam Jermyn, Tom Conerly, Nick Turner, Cem Anil, Carson Denison, Amanda Askell, Robert Lasenby, Yifan Wu, Shauna Kravec, Nicholas Schiefer, Tim Maxwell, Nicholas Joseph, Zac Hatfield-Dodds, Alex Tamkin, Karina Nguyen, Brayden McLean, Josiah… find
  12. Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeffrey Wu, Clemens Winter,… — Language Models are Few-Shot Learners
  13. Yuanpu Cao, Tianrong Zhang, Bochuan Cao, Ziyi Yin, Lu Lin, Fenglong Ma, and Jinghui Chen. 2024. Personalized steering of large language models: Versatile steering vectors through bi-directional preference optimization. In Advances in Neural Information Processing Systems 38: Annual Conference on Neu… find
  14. Huajun Chen. 2024. Large knowledge model: Perspectives and challenges. DATA INTELLIGENCE, 6(3):587–620. find
  15. Kai Chen, Zihao He, Taiwei Shi, and Kristina Lerman. 2025a. Steer-bench: A benchmark for evaluating the steerability of large language models. arXiv preprint arXiv:2505.20645. find
  16. Runjin Chen, Zhenyu Zhang, Junyuan Hong, Souvik Kundu, and Zhangyang Wang. 2025b. SEAL: steerable reasoning calibration of large language models for free. CoRR, abs/2504.07986. — [arXiv:2504.07986](https://arxiv.org/abs/2504.07986) find
  17. Yann Dubois, Balázs Galambosi, Percy Liang, and Tatsunori B. Hashimoto. 2024. Length-controlled alpacaeval: A simple way to debias automatic evaluators. CoRR, abs/2404.04475. find
  18. Esin Durmus, Alex Tamkin, Jack Clark, Jerry Wei, Jonathan Marcus, Joshua Batson, Kunal Handa, Liane Lovitt, Meg Tong, Miles McCain, Kunal Handa, Liane Lovitt, Meg Tong, Miles McCain, Oliver Rausch, Saffron Huang, Sam Bowman, Stuart Ritchie, Tom Henighan, and Deep Ganguli. 2024. Evaluating feature st… find
  19. Misbah Farooq, Varuna De Silva, Rahul Rahulamatha-van, and Xiyu Shi. 2025. Sentiment steering in large language models via activation vector manipulation. In DSP, pages 1–5. IEEE. find
  20. Leo Gao, Tom Dupré la Tour, Henk Tillman, Gabriel Goh, Rajan Troll, Alec Radford, Ilya Sutskever, Jan Leike, and Jeffrey Wu. 2024. Scaling and evaluating sparse autoencoders. CoRR, abs/2406.04093. find
  21. Chi Han, Jialiang Xu, Manling Li, Yi Fung, Chenkai Sun, Nan Jiang, Tarek F. Abdelzaher, and Heng Ji. 2024. Word embeddings are steers for language models. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), ACL 2024, Bangkok, Thailand,…
  22. Peixuan Han, Cheng Qian, Xiusi Chen, Yuji Zhang, Denghui Zhang, and Heng Ji. 2025. Internal activation as the polar star for steering unsafe LLM behavior. CoRR, abs/2502.01042. — [arXiv:2502.01042](https://arxiv.org/abs/2502.01042) find
  23. Robert Huben, Hoagy Cunningham, Logan Riggs, Aidan Ewart, and Lee Sharkey. 2024. Sparse autoencoders find highly interpretable features in language models. In The Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024. OpenReview.net. find
  24. Shawn Im and Yixuan Li. 2025. A unified understanding and evaluation of steering methods. CoRR, abs/2502.02716. — [arXiv:2502.02716](https://arxiv.org/abs/2502.02716) find
  25. Kenneth Li, Oam Patel, Fernanda B. Viégas, Hanspeter Pfister, and Martin Wattenberg. 2023. Inference-time intervention: Eliciting truthful answers from a language model. In Advances in Neural Information Processing Systems 36: Annual Conference on Neural Information Processing Systems 2023, NeurIPS…
  26. Tom Lieberum, Senthooran Rajamanoharan, Arthur Conmy, Lewis Smith, Nicolas Sonnerat, Vikrant Varma, János Kramár, Anca D. Dragan, Rohin Shah, and Neel Nanda. 2024. Gemma scope: Open sparse autoencoders everywhere all at once on gemma 2. CoRR, abs/2408.05147.
  27. Yitian Luo, Yu Liu, Lu Zhang, Feng Gao, and Jinguang Gu. 2025. A survey on quality evaluation of instruction fine-tuning datasets for large language models. DATA INTELLIGENCE, 7(3):527–566. find
  28. Aleksandar Makelov. 2024. Sparse autoencoders match supervised features for model steering on the ioi task. In ICML 2024 Workshop on Mechanistic Interpretability. find
  29. Samuel Marks and Max Tegmark. 2023. The geometry of truth: Emergent linear structure in large language model representations of true/false datasets. CoRR, abs/2310.06824.
  30. David Marr. 1982. Vision: A computational investigation into the human representation and processing of visual information. MIT press. find
  31. Tomás Mikolov, Wen-tau Yih, and Geoffrey Zweig. 2013. Linguistic regularities in continuous space word representations. In Human Language Technologies: Conference of the North American Chapter of the Association of Computational Linguistics, Proceedings, June 9-14, 2013, Westin Peachtree Plaza Hotel…
  32. George A Miller, Galanter Eugene, and Karl H Pribram. 2017. Plans and the structure of behaviour. In Systems research for behavioral science, pages 369–382. Routledge. find
  33. 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, EMNLP 2022, Abu Dhabi…
  34. Kiho Park, Todd Nief, Yo Joong Choe, and Victor Veitch. 2026. The information geometry of softmax: Probing and steering. Preprint, arXiv:2602.15293. — [arXiv:2602.15293](https://arxiv.org/abs/2602.15293) find
  35. Jeffrey Pennington, Richard Socher, and Christopher D. Manning. 2014. Glove: Global vectors for word representation. In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing, EMNLP 2014, October 25-29, 2014, Doha, Qatar, A meeting of SIGDAT, a Special Interest Group…
  36. Ethan Perez, Sam Ringer, Kamile Lukosiute, Karina Nguyen, Edwin Chen, Scott Heiner, Craig Pettit, Catherine Olsson, Sandipan Kundu, Saurav Kadavath, Andy Jones, Anna Chen, Benjamin Mann, Brian Israel, Bryan Seethor, Cameron McKinnon, Christopher Olah, Da Yan, Daniela Amodei, Dario Amodei, Dawn Drain… — Discovering Language Model Behaviors with Model-Written Evaluations
  37. Itamar Pres, Laura Ruis, Ekdeep Singh Lubana, and David Krueger. 2024. Towards reliable evaluation of behavior steering interventions in llms. CoRR, abs/2410.17245. — [arXiv:2410.17245](https://arxiv.org/abs/2410.17245) find
  38. Nina Rimsky, Nick Gabrieli, Julian Schulz, Meg Tong, Evan Hubinger, and Alexander Matt Turner. 2024. Steering llama 2 via contrastive activation addition. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), ACL 2024, Bangkok, Thailand,… find
  39. Basavaraj Sangayya Hiremath, Marco Polignano, Marco Levantesi, Giovanni Semeraro, Ernesto William De Luca, and Amos Poznanski. 2026. State-wise linear modulation (slim): A novel approach for steering large language models. Neural Networks, 199:108708. find
  40. Raphaël Sarfati, Eric Bigelow, Daniel Wurgaft, Jack Merullo, Atticus Geiger, Owen Lewis, Tom McGrath, and Ekdeep Singh Lubana. 2026. The shape of beliefs: Geometry, dynamics, and interventions along representation manifolds of language models’ posteriors. Preprint, arXiv:2602.02315. — [arXiv:2602.02315](https://arxiv.org/abs/2602.02315) find
  41. Lee Sharkey, Bilal Chughtai, Joshua Batson, Jack Lindsey, Jeff Wu, Lucius Bushnaq, Nicholas Goldowsky-Dill, Stefan Heimersheim, Alejandro Ortega, Joseph Bloom, et al. 2025. Open problems in mechanistic interpretability. arXiv preprint arXiv:2501.16496. — [arXiv:2501.16496](https://arxiv.org/abs/2501.16496) find
  42. Leheng Sheng, Changshuo Shen, Weixiang Zhao, Junfeng Fang, Xiaohao Liu, Zhenkai Liang, Xiang Wang, An Zhang, and Tat-Seng Chua. 2025. Alphasteer: Learning refusal steering with principled null-space constraint. CoRR, abs/2506.07022. — [arXiv:2506.07022](https://arxiv.org/abs/2506.07022) find
  43. Dong Shu, Xuansheng Wu, Haiyan Zhao, Daking Rai, Ziyu Yao, Ninghao Liu, and Mengnan Du. 2025. A survey on sparse autoencoders: Interpreting the internal mechanisms of large language models. In Findings of the Association for Computational Linguistics: EMNLP 2025, Suzhou, China, November 4-9, 2025, p… find
  44. Vincent Siu, Nicholas Crispino, David Park, Nathan W. Henry, Zhun Wang, Yang Liu, Dawn Song, and Chenguang Wang. 2025. Steeringsafety: A systematic safety evaluation framework of representation steering in llms. find
  45. Jiuding Sun, Sidharth Baskaran, Zhengxuan Wu, Michael Sklar, Christopher Potts, and Atticus Geiger. 2025. Hypersteer: Activation steering at scale with hypernetworks. CoRR, abs/2506.03292. find
  46. Magamed Taimeskhanov, Samuel Vaiter, and Damien Garreau. 2026. Towards understanding steering strength. Preprint, arXiv:2602.02712. — [arXiv:2602.02712](https://arxiv.org/abs/2602.02712) find
  47. Ala N. Tak, Amin Banayeeanzade, Anahita Bolourani, Mina Kian, Robin Jia, and Jonathan Gratch. 2025. Mechanistic interpretability of emotion inference in large language models. In Findings of the Association for Computational Linguistics, ACL 2025, Vienna, Austria, July 27 - August 1, 2025, pages 130… find
  48. Gemma Team. 2024a. Gemma: Open models based on gemini research and technology. CoRR, abs/2403.08295. find
  49. Llama Team. 2024b. The llama 3 herd of models. CoRR, abs/2407.21783.
  50. Qwen Team. 2024c. Qwen2.5: A party of foundation models. find
  51. Adly Templeton, Tom Conerly, Jonathan Marcus, Jack Lindsey, Trenton Bricken, Brian Chen, Adam Pearce, Craig Citro, Emmanuel Ameisen, Andy Jones, Hoagy Cunningham, Nicholas L Turner, Callum McDougall, Monte MacDiarmid, C. Daniel Freeman, Theodore R. Sumers, Edward Rees, Joshua Batson, Adam Jermyn, Sh… find
  52. Curt Tigges, Curt Tigges, Oskar Hollinsworth, Curt Tigges, Atticus Geiger, Atticus Geiger, Oskar Hollinsworth, Neel Nanda, Neel Nanda, Atticus Geiger, and Neel Nanda. 2023. Linear representations of sentiment in large language models. http://arxiv.org/abs/2310.15154. ↗
  53. Alexander Matt Turner, Lisa Thiergart, David Udell, Gavin Leech, Ulisse Mini, and Monte MacDiarmid. 2023. Activation addition: Steering language models without optimization. CoRR, abs/2308.10248.
  54. Mengru Wang, Ziwen Xu, Shengyu Mao, Shumin Deng, Zhaopeng Tu, Huajun Chen, and Ningyu Zhang. 2025. Beyond prompt engineering: Robust behavior control in llms via steering target atoms. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers),…
  55. Lyucheng Wu, Mengru Wang, Ziwen Xu, Tri Cao, Nay Oo, Bryan Hooi, and Shumin Deng. 2025a. Automating steering for safe multimodal large language models. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, EMNLP 2025, Suzhou, China, November 4-9, 2025, pages 792–… find
  56. Zhengxuan Wu, Aryaman Arora, Atticus Geiger, Zheng Wang, Jing Huang, Dan Jurafsky, Christopher D. Manning, and Christopher Potts. 2025b. Axbench: Steering llms? even simple baselines outperform sparse autoencoders. In Forty-second International Conference on Machine Learning, ICML 2025, Vancouver, B… find
  57. Zhengxuan Wu, Aryaman Arora, Zheng Wang, Atticus Geiger, Dan Jurafsky, Christopher D. Manning, and Christopher Potts. 2024. Reft: Representation fine-tuning for language models. In Advances in Neural Information Processing Systems 38: Annual Conference on Neural Information Processing Systems 2024,… find
  58. Zhengxuan Wu, Qinan Yu, Aryaman Arora, Christopher D. Manning, and Christopher Potts. 2025c. Improved representation steering for language models. CoRR, abs/2505.20809. find
  59. Ziwen Xu, Shuxun Wang, Kewei Xu, Haoming Xu, Mengru Wang, Xinle Deng, Yunzhi Yao, Guozhou Zheng, Huajun Chen, and Ningyu Zhang. 2025. Easyedit2: An easy-to-use steering framework for editing large language models. CoRR, abs/2504.15133. find
  60. Ziwen Xu, Chenyan Wu, Hengyu Sun, Haiwen Hong, Mengru Wang, Yunzhi Yao, Longtao Huang, Hui Xue, Shumin Deng, Zhixuan Chu, Huajun Chen, and Ningyu Zhang. 2026. Why steering works: Toward a unified view of language model parameter dynamics. CoRR, abs/2602.02343.
  61. Hengyuan Zhang, Zhihao Zhang, Mingyang Wang, Zun-hai Su, Yiwei Wang, Qianli Wang, Shuzhou Yuan, Ercong Nie, Xufeng Duan, Qibo Xue, Zeping Yu, Chenming Shang, Xiao Liang, Jing Xiong, Hui Shen, Chaofan Tao, Zhengwu Liu, Senjie Jin, Zhiheng Xi, Dongdong Zhang, Sophia Ananiadou, Tao Gui, Ruobing Xie, Ha… — Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models
  62. Jinghao Zhang, Yuting Liu, Wenjie Wang, Qiang Liu, Shu Wu, Liang Wang, and Tat-Seng Chua. 2025. Personalized text generation with contrastive activation steering. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), ACL 2025, Vienna, Aus…
  63. Detailed experimental results for all domain across three granularity levels (L1–L3) can be found in Table 7, 8, 9, 10. We report the Concept Score (CS), Instruction Score (IS), Fluency Score (FS), and their Harmonic Mean (HM). All metrics are evaluated on a 0–4 scale. find
  64. Wayne Xin Zhao, Kun Zhou, Junyi Li, Tianyi Tang, Xiaolei Wang, Yupeng Hou, Yingqian Min, Beichen Zhang, Junjie Zhang, Zican Dong, Yifan Du, Chen Yang, Yushuo Chen, Zhipeng Chen, Jinhao Jiang, Ruiyang Ren, Yifan Li, Xinyu Tang, Zikang Liu, Peiyu Liu, Jian-Yun Nie, and Ji-Rong Wen. 2023. A survey of l…
  65. Minjun Zhu, Yixuan Weng, Linyi Yang, and Yue Zhang. 2025. Personality alignment of large language models. In The Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24-28, 2025. OpenReview.net.
  66. Andy Zou, Long Phan, Sarah Li Chen, James Campbell, Phillip Guo, Richard Ren, Alexander Pan, Xuwang Yin, Mantas Mazeika, Ann-Kathrin Dombrowski, Shashwat Goel, Nathaniel Li, Michael J. Byun, Zifan Wang, Alex Mallen, Steven Basart, Sanmi Koyejo, Dawn Song, Matt Fredrikson, J. Zico Kolter, and Dan Hen… — Representation Engineering: A Top-Down Approach to AI Transparency

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/xu-2026-how-controllable-are-large-language/