{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,7,6]],"date-time":"2026-07-06T18:20:52Z","timestamp":1783362052862,"version":"3.54.6"},"reference-count":435,"publisher":"Elsevier BV","license":[{"start":{"date-parts":[[2026,11,1]],"date-time":"2026-11-01T00:00:00Z","timestamp":1793491200000},"content-version":"tdm","delay-in-days":0,"URL":"https:\/\/www.elsevier.com\/tdm\/userlicense\/1.0\/"},{"start":{"date-parts":[[2026,11,1]],"date-time":"2026-11-01T00:00:00Z","timestamp":1793491200000},"content-version":"tdm","delay-in-days":0,"URL":"https:\/\/www.elsevier.com\/legal\/tdmrep-license"},{"start":{"date-parts":[[2026,11,1]],"date-time":"2026-11-01T00:00:00Z","timestamp":1793491200000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-017"},{"start":{"date-parts":[[2026,11,1]],"date-time":"2026-11-01T00:00:00Z","timestamp":1793491200000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-037"},{"start":{"date-parts":[[2026,11,1]],"date-time":"2026-11-01T00:00:00Z","timestamp":1793491200000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-012"},{"start":{"date-parts":[[2026,11,1]],"date-time":"2026-11-01T00:00:00Z","timestamp":1793491200000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-029"},{"start":{"date-parts":[[2026,11,1]],"date-time":"2026-11-01T00:00:00Z","timestamp":1793491200000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-004"}],"funder":[{"DOI":"10.13039\/501100003803","name":"The University of Hong Kong","doi-asserted-by":"publisher","id":[{"id":"10.13039\/501100003803","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100013105","name":"Shanghai Rising-Star Program","doi-asserted-by":"publisher","id":[{"id":"10.13039\/501100013105","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/100017948","name":"Chongqing Academy of Science and Technology","doi-asserted-by":"publisher","id":[{"id":"10.13039\/100017948","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100002920","name":"Research Grants Council, University Grants Committee","doi-asserted-by":"publisher","id":[{"id":"10.13039\/501100002920","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":["elsevier.com","sciencedirect.com"],"crossmark-restriction":true},"short-container-title":["Computer Science Review"],"published-print":{"date-parts":[[2026,11]]},"DOI":"10.1016\/j.cosrev.2026.101011","type":"journal-article","created":{"date-parts":[[2026,6,11]],"date-time":"2026-06-11T23:40:56Z","timestamp":1781221256000},"page":"101011","update-policy":"https:\/\/doi.org\/10.1016\/elsevier_cm_policy","source":"Crossref","is-referenced-by-count":0,"special_numbering":"C","title":["Locate, steer, and improve: A practical survey of actionable mechanistic interpretability in large language models"],"prefix":"10.1016","volume":"62","author":[{"given":"Hengyuan","family":"Zhang","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Zhihao","family":"Zhang","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Mingyang","family":"Wang","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Zunhai","family":"Su","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yiwei","family":"Wang","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Qianli","family":"Wang","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Shuzhou","family":"Yuan","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Ercong","family":"Nie","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Xufeng","family":"Duan","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Qibo","family":"Xue","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Zeping","family":"Yu","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Chenming","family":"Shang","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Xiao","family":"Liang","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Jing","family":"Xiong","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Hui","family":"Shen","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Chaofan","family":"Tao","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Zhengwu","family":"Liu","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Senjie","family":"Jin","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Zhiheng","family":"Xi","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Dongdong","family":"Zhang","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Sophia","family":"Ananiadou","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Tao","family":"Gui","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Ruobing","family":"Xie","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Hayden Kwok-Hay","family":"So","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Hinrich","family":"Sch\u00fctze","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Xuanjing","family":"Huang","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Qi","family":"Zhang","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Ngai","family":"Wong","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"78","reference":[{"key":"10.1016\/j.cosrev.2026.101011_bib0005","series-title":"Advances in Neural Information Processing Systems (NeurIPS)","article-title":"Sanity checks for saliency maps","author":"Adebayo","year":"2018"},{"key":"10.1016\/j.cosrev.2026.101011_bib0010","author":"Ahsan"},{"key":"10.1016\/j.cosrev.2026.101011_bib0015","series-title":"48th IEEE Annual Computers, Software, and Applications Conference, COMPSAC 2024, Osaka, Japan, July 2\u20134, 2024","first-page":"1057","article-title":"Uncovering the interpretation of large language models","author":"Akter","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0020","series-title":"5th International Conference on Learning Representations, ICLR 2017, Toulon, France, April 24\u201326, 2017, Workshop Track Proceedings","article-title":"Understanding intermediate layers using linear classifier probes","author":"Alain","year":"2017"},{"key":"10.1016\/j.cosrev.2026.101011_bib0025","series-title":"Advances in Neural Information Processing Systems 35: Annual Conference on Neural Information Processing Systems 2022, NeurIPS 2022, New Orleans, LA, USA, November 28 - December 9, 2022","article-title":"Flamingo: a visual language model for few-shot learning","author":"Alayrac","year":"2022"},{"key":"10.1016\/j.cosrev.2026.101011_bib0030","author":"Allen-Zhu"},{"key":"10.1016\/j.cosrev.2026.101011_bib0035","article-title":"Circuit tracing: revealing computational graphs in language models","author":"Ameisen","year":"2025","journal-title":"Transformer Circuits Thread"},{"key":"10.1016\/j.cosrev.2026.101011_bib0040","author":"An"},{"key":"10.1016\/j.cosrev.2026.101011_bib0045","author":"Andrylie"},{"key":"10.1016\/j.cosrev.2026.101011_bib0050","author":"Arad"},{"key":"10.1016\/j.cosrev.2026.101011_bib0055","series-title":"The Thirty-Eighth Annual Conference on Neural Information Processing Systems","article-title":"Refusal in language models is mediated by a single direction","author":"Arditi","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0060","doi-asserted-by":"crossref","first-page":"100213","DOI":"10.52202\/079017-3180","article-title":"Quarot: outlier-free 4-bit inference in rotated LLMs","volume":"37","author":"Ashkboos","year":"2024","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"10.1016\/j.cosrev.2026.101011_bib0065","series-title":"Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), ACL 2025, Vienna, Austria, July 27 - August 1, 2025","first-page":"19954","article-title":"Generalized attention flow: feature attribution for transformer models via maximum flow","author":"Azarkhalili","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0070","author":"Ball"},{"key":"10.1016\/j.cosrev.2026.101011_bib0075","author":"Banayeeanzade"},{"key":"10.1016\/j.cosrev.2026.101011_bib0080","author":"Bas"},{"key":"10.1016\/j.cosrev.2026.101011_bib0085","author":"Bayat"},{"key":"10.1016\/j.cosrev.2026.101011_bib0090","doi-asserted-by":"crossref","first-page":"207","DOI":"10.1162\/coli_a_00422","article-title":"Probing classifiers: promises, shortcomings, and advances","volume":"48","author":"Belinkov","year":"2022","journal-title":"Comput. Linguistics"},{"key":"10.1016\/j.cosrev.2026.101011_bib0095","author":"Belrose"},{"key":"10.1016\/j.cosrev.2026.101011_bib0100","author":"Bereska"},{"key":"10.1016\/j.cosrev.2026.101011_bib0105","series-title":"Proceedings of the Computer Vision and Pattern Recognition Conference","first-page":"4135","article-title":"Unveiling visual perception in language models: an attention head analysis approach","author":"Bi","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0110","series-title":"Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, EMNLP 2024, Miami, FL, USA, November 12\u201316, 2024","first-page":"14113","article-title":"Hopping too late: exploring the limitations of large language models on multi-hop queries","author":"Biran","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0115","author":"Bloom"},{"key":"10.1016\/j.cosrev.2026.101011_bib0120","series-title":"Advances in Neural Information Processing Systems","first-page":"75067","article-title":"Quantizable transformers: removing outliers by helping attention heads do nothing","author":"Bondarenko","year":"2023"},{"key":"10.1016\/j.cosrev.2026.101011_bib0125","author":"Braun"},{"key":"10.1016\/j.cosrev.2026.101011_bib0130","first-page":"15301","article-title":"Attention approximates sparse distributed memory","volume":"34","author":"Bricken","year":"2021","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"10.1016\/j.cosrev.2026.101011_bib0135","article-title":"Towards monosemanticity: decomposing language models with dictionary learning","author":"Bricken","year":"2023","journal-title":"Transformer Circuits Thread"},{"key":"10.1016\/j.cosrev.2026.101011_bib0140","series-title":"Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)","first-page":"6131","article-title":"Large language models share representations of latent grammatical concepts across typologically diverse languages","author":"Brinkmann","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0145","author":"Bussmann"},{"key":"10.1016\/j.cosrev.2026.101011_bib0150","author":"Cai"},{"key":"10.1016\/j.cosrev.2026.101011_bib0155","author":"Cai"},{"key":"10.1016\/j.cosrev.2026.101011_bib0160","series-title":"Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","first-page":"4792","article-title":"Spectral filters, dark signals, and attention sinks","author":"Cancedda","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0165","author":"Chandna"},{"key":"10.1016\/j.cosrev.2026.101011_bib0170","author":"Chang"},{"key":"10.1016\/j.cosrev.2026.101011_bib0175","author":"Chanin"},{"key":"10.1016\/j.cosrev.2026.101011_bib0180","series-title":"The Thirty-Ninth Annual Conference on Neural Information Processing Systems","article-title":"Transferring linear features across language models with model stitching","author":"Chen","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0185","author":"Chen"},{"key":"10.1016\/j.cosrev.2026.101011_bib0190","series-title":"AAAI-25, Sponsored by the Association for the Advancement of Artificial Intelligence, February 25 - March 4, 2025, Philadelphia, PA, USA","first-page":"23595","article-title":"Identifying query-relevant neurons in large language models for long-form texts","author":"Chen","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0195","author":"Chen"},{"key":"10.1016\/j.cosrev.2026.101011_bib0200","series-title":"Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers)","first-page":"256","article-title":"Learnable privacy neurons localization in language models","author":"Chen","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0205","series-title":"Forty-First International Conference on Machine Learning","article-title":"In-context sharpness as alerts: an inner representation perspective for hallucination mitigation","author":"Chen","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0210","series-title":"Forty-First International Conference on Machine Learning","article-title":"From Yes-men to truth-tellers: addressing sycophancy in large language models with pinpoint tuning","author":"Chen","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0215","series-title":"The Fourteenth International Conference on Learning Representations","article-title":"Neuron-aware data selection in instruction tuning for large language models","author":"Chen","year":"2026"},{"key":"10.1016\/j.cosrev.2026.101011_bib0220","first-page":"17817","article-title":"Journey to the center of the knowledge neurons: discoveries of language-independent knowledge neurons and degenerate knowledge neurons","author":"Chen","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0225","series-title":"The Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24\u201328, 2025","article-title":"Knowledge localization: mission not accomplished? Enter query localization!","author":"Chen","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0230","series-title":"Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","first-page":"28241","article-title":"Inner thinking transformer: leveraging dynamic depth scaling to foster adaptive internal thinking","author":"Chen","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0235","series-title":"Findings of the Association for Computational Linguistics: ACL 2023","first-page":"12572","article-title":"Can cross-lingual transferability of multilingual transformers be activated without end-task data?","author":"Chi","year":"2023"},{"key":"10.1016\/j.cosrev.2026.101011_bib0240","series-title":"Proceedings of the 6th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP","first-page":"379","article-title":"Identifying and adapting transformer-components responsible for gender bias in an english language model","author":"Chintam","year":"2023"},{"key":"10.1016\/j.cosrev.2026.101011_bib0245","author":"Cho"},{"key":"10.1016\/j.cosrev.2026.101011_bib0250","series-title":"Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing","first-page":"28949","article-title":"Toward efficient sparse autoencoder-guided steering for improved in-context learning in large language models","author":"Cho","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0255","series-title":"The Twelfth International Conference on Learning Representations","article-title":"DoLa: decoding by contrasting layers improves factuality in large language models","author":"Chuang","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0260","author":"Conklin"},{"key":"10.1016\/j.cosrev.2026.101011_bib0265","series-title":"Advances in Neural Information Processing Systems","first-page":"16318","article-title":"Towards automated circuit discovery for mechanistic interpretability","author":"Conmy","year":"2023"},{"key":"10.1016\/j.cosrev.2026.101011_bib0270","series-title":"Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics, ACL 2018, Melbourne, Australia, July 15\u201320, 2018, Volume 1: Long Papers","first-page":"2126","article-title":"What you can cram into a single $&!#* vector: probing sentence embeddings for linguistic properties","author":"Conneau","year":"2018"},{"key":"10.1016\/j.cosrev.2026.101011_bib0275","author":"Cui"},{"key":"10.1016\/j.cosrev.2026.101011_bib0280","author":"Cunningham"},{"key":"10.1016\/j.cosrev.2026.101011_bib0285","author":"Cywi\u0144ski"},{"key":"10.1016\/j.cosrev.2026.101011_bib0290","author":"Da Silva"},{"key":"10.1016\/j.cosrev.2026.101011_bib0295","series-title":"Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), ACL 2022, Dublin, Ireland, May 22\u201327, 2022","first-page":"8493","article-title":"Knowledge neurons in pretrained transformers","author":"Dai","year":"2022"},{"key":"10.1016\/j.cosrev.2026.101011_bib0300","author":"Davies"},{"key":"10.1016\/j.cosrev.2026.101011_bib0305","series-title":"The Thirteenth International Conference on Learning Representations","article-title":"Neuron based personality trait induction in large language models","author":"Deng","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0310","doi-asserted-by":"crossref","first-page":"30318","DOI":"10.52202\/068431-2198","article-title":"GPT3. Int8 (\u00a0): 8-bit matrix multiplication for transformers at scale","volume":"35","author":"Dettmers","year":"2022","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"10.1016\/j.cosrev.2026.101011_bib0315","series-title":"Proceedings of the 6th ACM International Conference on AI in Finance","first-page":"96","article-title":"Tracing positional bias in financial decision-making: mechanistic insights from Qwen2.5","author":"Dimino","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0320","author":"Dreyer"},{"key":"10.1016\/j.cosrev.2026.101011_bib0325","author":"Du"},{"key":"10.1016\/j.cosrev.2026.101011_bib0330","series-title":"Proceedings of the 31st International Conference on Computational Linguistics","first-page":"10148","article-title":"Unveiling language competence neurons: a psycholinguistic approach to model interpretability","author":"Duan","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0335","unstructured":"A. Dubey, A. Jauhri, A. Pandey, A. Kadian, A. Al-Dahle, A. Letman, A. Mathur, A. Schelten, A. Yang, A. Fan, et al., The LLaMA 3 herd of models, 2024. arXiv e-prints, arXiv\u20132407."},{"key":"10.1016\/j.cosrev.2026.101011_bib0340","author":"Dumitru"},{"key":"10.1016\/j.cosrev.2026.101011_bib0345","author":"Elhage"},{"key":"10.1016\/j.cosrev.2026.101011_bib0350","article-title":"A mathematical framework for transformer circuits","author":"Elhage","year":"2021","journal-title":"Transformer Circuits Thread"},{"key":"10.1016\/j.cosrev.2026.101011_bib0355","series-title":"Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","first-page":"12622","article-title":"LayerSkip: enabling early exit inference and self-speculative decoding","author":"Elhoushi","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0360","series-title":"Findings of the Association for Computational Linguistics: ACL 2023, Toronto, Canada, July 9\u201314, 2023","first-page":"7555","article-title":"Sequential integrated gradients: a simple but effective method for explaining language models","author":"Enguehard","year":"2023"},{"key":"10.1016\/j.cosrev.2026.101011_bib0365","first-page":":120:1","article-title":"Switch transformers: scaling to trillion parameter models with simple and efficient sparsity","volume":"23","author":"Fedus","year":"2022","journal-title":"J. Mach. Learn. Res."},{"key":"10.1016\/j.cosrev.2026.101011_bib0370","series-title":"NeurIPS 2023 Workshop on Symmetry and Geometry in Neural Representations","article-title":"How do language models bind entities in context?","author":"Feng","year":"2023"},{"key":"10.1016\/j.cosrev.2026.101011_bib0375","author":"Ferrando"},{"key":"10.1016\/j.cosrev.2026.101011_bib0380","series-title":"Workshop on Actionable Interpretability at ICML 2025","article-title":"Truthful or fabricated? Using causal attribution to mitigate reward hacking in explanations","author":"Ferreira","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0385","article-title":"Towards empirical interpretation of internal circuits and properties in grokked transformers on modular polynomials","author":"Furuta","year":"2024","journal-title":"Trans. Mach. Learn. Res."},{"key":"10.1016\/j.cosrev.2026.101011_bib0390","author":"Galichin"},{"key":"10.1016\/j.cosrev.2026.101011_bib0395","series-title":"2025 International Conference on Data Science, Agents & Artificial Intelligence (ICDSAAI)","first-page":"1","article-title":"Exploring mechanistic interpretability in large language models: challenges, approaches, and insights","author":"Gantla","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0400","author":"Gao"},{"key":"10.1016\/j.cosrev.2026.101011_bib0405","author":"Gao"},{"key":"10.1016\/j.cosrev.2026.101011_bib0410","author":"Gao"},{"key":"10.1016\/j.cosrev.2026.101011_bib0415","first-page":"1","article-title":"Causal abstraction: a theoretical foundation for mechanistic interpretability","volume":"26","author":"Geiger","year":"2025","journal-title":"J. Mach. Learn. Res."},{"key":"10.1016\/j.cosrev.2026.101011_bib0420","series-title":"Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing","first-page":"12216","article-title":"Dissecting recall of factual associations in auto-regressive language models","author":"Geva","year":"2023"},{"key":"10.1016\/j.cosrev.2026.101011_bib0425","series-title":"Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing","first-page":"30","article-title":"Transformer feed-forward layers build predictions by promoting concepts in the vocabulary space","author":"Geva","year":"2022"},{"key":"10.1016\/j.cosrev.2026.101011_bib0430","series-title":"Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing","first-page":"5484","article-title":"Transformer feed-forward layers are key-value memories","author":"Geva","year":"2021"},{"key":"10.1016\/j.cosrev.2026.101011_bib0435","series-title":"Advances in Neural Information Processing Systems","first-page":"125967","article-title":"Who\u2019 s asking? User personas and the mechanics of latent misalignment","author":"Ghandeharioun","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0440","author":"Ghilardi"},{"key":"10.1016\/j.cosrev.2026.101011_bib0445","author":"Goldowsky-Dill"},{"key":"10.1016\/j.cosrev.2026.101011_bib0450","series-title":"Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing","first-page":"12702","article-title":"Breaking bad tokens: detoxification of LLMs using sparse autoencoders","author":"Goyal","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0455","series-title":"Proceedings of the European Chapter of the Association for Computational Linguistics (EACL)","article-title":"Steering safely or off a cliff? Rethinking specificity and robustness in inference-time interventions","author":"Goyal","year":"2026"},{"key":"10.1016\/j.cosrev.2026.101011_bib0460","doi-asserted-by":"crossref","first-page":"105","DOI":"10.1016\/j.actpsy.2003.12.003","article-title":"Executive control emerging from dynamic interactions between brain systems mediating language, working memory and attentional processes","volume":"115","author":"Gruber","year":"2004","journal-title":"Acta Psychol."},{"key":"10.1016\/j.cosrev.2026.101011_bib0465","series-title":"Handbook of Individual Differences in Cognition: Attention, Memory, and Executive Control","author":"Gruszka","year":"2010"},{"key":"10.1016\/j.cosrev.2026.101011_bib0470","author":"Guan"},{"key":"10.1016\/j.cosrev.2026.101011_bib0475","series-title":"Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing","first-page":"21158","article-title":"Attention score is not all you need for token importance indicator in KV cache reduction: value also matters","author":"Guo","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0480","author":"Gur-Arieh"},{"key":"10.1016\/j.cosrev.2026.101011_bib0485","author":"Gurgurov"},{"key":"10.1016\/j.cosrev.2026.101011_bib0490","author":"Gurgurov"},{"key":"10.1016\/j.cosrev.2026.101011_bib0495","series-title":"Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), ACL 2025, Vienna, Austria, July 27 - August 1, 2025","first-page":"2792","article-title":"Position-aware automatic circuit discovery","author":"Haklay","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0500","series-title":"ICLR 2026","article-title":"ZeroTuning: unlocking the initial token\u2019s power to enhance large language models without training","author":"Han","year":"2026"},{"key":"10.1016\/j.cosrev.2026.101011_bib0505","series-title":"NeurIPS 2025 Workshop on Evaluating the Evolving LLM Lifecycle: Benchmarks, Emergent Abilities, and Scaling","article-title":"Personality as a probe for LLM evaluation: method trade-offs and downstream effects","author":"Handa","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0510","series-title":"Advances in Neural Information Processing Systems 36: Annual Conference on Neural Information Processing Systems 2023, NeurIPS 2023, New Orleans, LA, USA, December 10 - 16, 2023","article-title":"How does GPT-2 compute greater-than?: Interpreting mathematical abilities in a pre-trained language model","author":"Hanna","year":"2023"},{"key":"10.1016\/j.cosrev.2026.101011_bib0515","series-title":"First Conference on Language Modeling","article-title":"Have faith in faithfulness: going beyond circuit overlap when finding model mechanisms","author":"Hanna","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0520","series-title":"Proceedings of the 8th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP","first-page":"239","article-title":"Circuit-tracer: a new library for finding feature circuits","author":"Hanna","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0525","doi-asserted-by":"crossref","first-page":"68287","DOI":"10.52202\/079017-2181","article-title":"Zipcache: accurate and efficient kv cache quantization with salient token identification","volume":"37","author":"He","year":"2024","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"10.1016\/j.cosrev.2026.101011_bib0530","author":"He"},{"key":"10.1016\/j.cosrev.2026.101011_bib0535","author":"He"},{"key":"10.1016\/j.cosrev.2026.101011_bib0540","author":"He"},{"key":"10.1016\/j.cosrev.2026.101011_bib0545","series-title":"Forty-Second International Conference on Machine Learning","article-title":"To steer or not to steer? Mechanistic error reduction with abstention for language models","author":"Hedstr\u00f6m","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0550","series-title":"Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, EMNLP-IJCNLP 2019, Hong Kong, China, November 3\u20137, 2019","first-page":"2733","article-title":"Designing and interpreting probes with control tasks","author":"Hewitt","year":"2019"},{"key":"10.1016\/j.cosrev.2026.101011_bib0555","series-title":"Findings of the Association for Computational Linguistics: EMNLP 2024","first-page":"13402","article-title":"Why do LLaVA vision-language models reply to images in english?","author":"Hinck","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0560","series-title":"Findings of the Association for Computational Linguistics: ACL 2024","first-page":"14596","article-title":"Cyclealign: iterative distillation from black-box LLM to white-box models for better human alignment","author":"Hong","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0565","series-title":"Findings of the Association for Computational Linguistics: ACL 2025","first-page":"21565","article-title":"The reasoning-memorization interplay in language models is mediated by a single direction","author":"Hong","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0570","author":"Hoogland"},{"key":"10.1016\/j.cosrev.2026.101011_bib0575","series-title":"Advances in Neural Information Processing Systems","first-page":"1270","article-title":"KVQuant: towards 10 million context length LLM inference with KV cache quantization","author":"Hooper","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0580","series-title":"International Conference on Machine Learning, ICML 2023, 23\u201329 July 2023, Honolulu, Hawaii, USA","first-page":"13285","article-title":"Decoding layer saliency in language transformers","author":"Hou","year":"2023"},{"key":"10.1016\/j.cosrev.2026.101011_bib0585","author":"Hsueh"},{"key":"10.1016\/j.cosrev.2026.101011_bib0590","series-title":"International Conference on Learning Representations","article-title":"LoRA: low-rank adaptation of large language models","author":"Hu","year":"2022"},{"key":"10.1016\/j.cosrev.2026.101011_bib0595","author":"Hu"},{"key":"10.1016\/j.cosrev.2026.101011_bib0600","author":"Huang"},{"key":"10.1016\/j.cosrev.2026.101011_bib0605","series-title":"Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing","first-page":"15471","article-title":"Pierce the mists, greet the sky: decipher knowledge overshadowing via knowledge circuit analysis","author":"Huang","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0610","author":"Huang"},{"key":"10.1016\/j.cosrev.2026.101011_bib0615","series-title":"First Conference on Language Modeling","article-title":"Unified view of grokking, double descent and emergent abilities: a comprehensive study on algorithm task","author":"Huang","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0620","series-title":"The Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7\u201311, 2024","article-title":"Sparse autoencoders find highly interpretable features in language models","author":"Huben","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0625","series-title":"Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing","first-page":"6801","article-title":"MMNeuron: discovering neuron-level domain-specific interpretation in multimodal large language model","author":"Huo","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0630","author":"H\u00f8jer"},{"key":"10.1016\/j.cosrev.2026.101011_bib0635","series-title":"The Eleventh International Conference on Learning Representations","article-title":"Editing models with task arithmetic","author":"Ilharco","year":"2023"},{"key":"10.1016\/j.cosrev.2026.101011_bib0640","author":"Ismail"},{"key":"10.1016\/j.cosrev.2026.101011_bib0645","author":"Jafari"},{"key":"10.1016\/j.cosrev.2026.101011_bib0650","series-title":"Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)","first-page":"1041","article-title":"On large language models\u2019 hallucination with regard to known facts","author":"Jiang","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0655","unstructured":"G. Jiang, Z. Li, C. Jiang, S. Xue, J. Zhou, L. Song, D. Lian, Y. Wei, Interpretable catastrophic forgetting of large language model fine-tuning via instruction vector, 2024. arXiv e-prints, arXiv\u20132406."},{"key":"10.1016\/j.cosrev.2026.101011_bib0660","author":"Jiang"},{"key":"10.1016\/j.cosrev.2026.101011_bib0665","author":"Jiang"},{"key":"10.1016\/j.cosrev.2026.101011_bib0670","series-title":"Proceedings of the Computer Vision and Pattern Recognition Conference","first-page":"25004","article-title":"Devils in middle layers of large vision-language models: interpreting, detecting and mitigating object hallucinations via attention lens","author":"Jiang","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0675","series-title":"Forty-Second International Conference on Machine Learning, ICML 2025, Vancouver, BC, Canada, July 13\u201319, 2025","article-title":"Massive values in self-attention modules are the key to contextual knowledge understanding","author":"Jin","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0680","series-title":"Proceedings of the 31st International Conference on Computational Linguistics, COLING 2025, Abu Dhabi, UAE, January 19\u201324, 2025","first-page":"558","article-title":"Exploring concept depth: how large language models acquire knowledge and concept at different layers?","author":"Jin","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0685","series-title":"Findings of the Association for Computational Linguistics, ACL 2024, Bangkok, Thailand and Virtual Meeting, August 11\u201316, 2024","first-page":"1193","article-title":"Cutting off the head ends the conflict: a mechanism for interpreting and mitigating knowledge conflicts in language models","author":"Jin","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0690","series-title":"Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing","first-page":"28220","article-title":"LinguaLens: towards interpreting linguistic mechanisms of large language models via sparse auto-encoder","author":"Jing","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0695","series-title":"Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing","first-page":"6346","article-title":"Personas as a way to model truthfulness in language models","author":"Joshi","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0700","series-title":"Second Conference on Language Modeling","article-title":"Probing then editing response personality of large language models","author":"Ju","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0705","series-title":"Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation, LREC\/COLING 2024, 20\u201325 May, 2024, Torino, Italy","first-page":"8235","article-title":"How large language models encode context knowledge? A layer-wise probing study","author":"Ju","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0710","series-title":"Findings of the Association for Computational Linguistics: EMNLP 2023, Singapore, December 6\u201310, 2023","first-page":"7721","article-title":"Impact of co-occurrence on factual knowledge of large language models","author":"Kang","year":"2023"},{"key":"10.1016\/j.cosrev.2026.101011_bib0715","series-title":"Forty-Second International Conference on Machine Learning, ICML 2025, Vancouver, BC, Canada, July 13\u201319, 2025","article-title":"Are sparse autoencoders useful? A case study in sparse probing","author":"Kantamneni","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0720","author":"Kargaran"},{"key":"10.1016\/j.cosrev.2026.101011_bib0725","author":"Karny"},{"key":"10.1016\/j.cosrev.2026.101011_bib0730","author":"Karvonen"},{"key":"10.1016\/j.cosrev.2026.101011_bib0735","author":"Kassem"},{"key":"10.1016\/j.cosrev.2026.101011_bib0740","series-title":"Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, EMNLP 2024, Miami, FL, USA, November 12\u201316, 2024","first-page":"2390","article-title":"Backward lens: projecting language model gradients into the vocabulary space","author":"Katz","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0745","author":"Kendiukhov"},{"key":"10.1016\/j.cosrev.2026.101011_bib0750","author":"Kharlapenko"},{"key":"10.1016\/j.cosrev.2026.101011_bib0755","series-title":"The Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24\u201328, 2025","article-title":"Linear representations of political perspective emerge in large language models","author":"Kim","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0760","series-title":"The Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24\u201328, 2025","article-title":"Linear representations of political perspective emerge in large language models","author":"Kim","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0765","series-title":"Mechanistic Interpretability Workshop at NeurIPS 2025","article-title":"Interpreting attention heads for image-to-text information flow in large vision\u2013language models","author":"Kim","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0770","author":"Kim"},{"key":"10.1016\/j.cosrev.2026.101011_bib0775","series-title":"Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)","first-page":"6919","article-title":"On the multilingual ability of decoder-based pre-trained language models: finding and controlling language-specific neurons","author":"Kojima","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0780","series-title":"Forty-Second International Conference on Machine Learning, ICML 2025, Vancouver, BC, Canada, July 13\u201319, 2025","article-title":"Joint localization and activation editing for low-resource fine-tuning","author":"Lai","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0785","series-title":"Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing","first-page":"13427","article-title":"Style-specific neurons for steering LLMs in text style transfer","author":"Lai","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0790","author":"Laitenberger"},{"key":"10.1016\/j.cosrev.2026.101011_bib0795","author":"Lawson"},{"key":"10.1016\/j.cosrev.2026.101011_bib0800","series-title":"Forty-First International Conference on Machine Learning, ICML 2024, Vienna, Austria, July 21\u201327, 2024","article-title":"A mechanistic understanding of alignment algorithms: a case study on DPO and toxicity","author":"Lee","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0805","series-title":"Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing","first-page":"21514","article-title":"Interpretation meets safety: a survey on interpretation methods and tools for improving LLM safety","author":"Lee","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0810","author":"Lei"},{"key":"10.1016\/j.cosrev.2026.101011_bib0815","series-title":"Forty-Second International Conference on Machine Learning, ICML 2025, Vancouver, BC, Canada, July 13\u201319, 2025","article-title":"Taming knowledge conflicts in language models","author":"Li","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0820","series-title":"Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing","first-page":"31290","article-title":"LoRACoE: improving large language model via composition-based LoRA expert","author":"Li","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0825","series-title":"AAAI-25, Sponsored by the Association for the Advancement of Artificial Intelligence, February 25 - March 4, 2025, Philadelphia, PA, USA","first-page":"24431","article-title":"Enhancing large language model performance with gradient-based parameter selection","author":"Li","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0830","series-title":"NAACL HLT 2016, the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, San Diego California, USA, June 12\u201317, 2016","first-page":"681","article-title":"Visualizing and understanding neural models in NLP","author":"Li","year":"2016"},{"key":"10.1016\/j.cosrev.2026.101011_bib0835","series-title":"International Conference on Machine Learning, ICML 2023, 23\u201329 July 2023, Honolulu, Hawaii, USA","first-page":"19730","article-title":"BLIP-2: bootstrapping language-image pre-training with frozen image encoders and large language models","author":"Li","year":"2023"},{"key":"10.1016\/j.cosrev.2026.101011_bib0840","author":"Li"},{"key":"10.1016\/j.cosrev.2026.101011_bib0845","author":"Li"},{"key":"10.1016\/j.cosrev.2026.101011_bib0850","series-title":"Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","first-page":"32017","article-title":"What happened in LLMs layers when trained for fast vs. slow thinking: a gradient perspective","author":"Li","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0855","series-title":"The Fourteenth International Conference on Learning Representations","article-title":"Attributing response to context: a jensen-shannon divergence driven mechanistic study of context attribution in retrieval-augmented generation","author":"Li","year":"2026"},{"key":"10.1016\/j.cosrev.2026.101011_bib0860","author":"Li"},{"key":"10.1016\/j.cosrev.2026.101011_bib0865","author":"Li"},{"key":"10.1016\/j.cosrev.2026.101011_bib0870","series-title":"The Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24\u201328, 2025","article-title":"Safety layers in aligned large language models: the key to LLM security","author":"Li","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0875","author":"Li"},{"key":"10.1016\/j.cosrev.2026.101011_bib0880","author":"Li"},{"key":"10.1016\/j.cosrev.2026.101011_bib0885","author":"Li"},{"key":"10.1016\/j.cosrev.2026.101011_bib0890","author":"Li"},{"key":"10.1016\/j.cosrev.2026.101011_bib0895","series-title":"Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing","first-page":"10893","article-title":"Feature extraction and steering for enhanced chain-of-thought reasoning in language models","author":"Li","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0900","author":"Li"},{"key":"10.1016\/j.cosrev.2026.101011_bib0905","author":"Liang"},{"key":"10.1016\/j.cosrev.2026.101011_bib0910","series-title":"Findings of the Association for Computational Linguistics: EMNLP 2020","first-page":"1663","article-title":"On the language neutrality of pre-trained multilingual representations","author":"Libovick\u00fd","year":"2020"},{"key":"10.1016\/j.cosrev.2026.101011_bib0915","series-title":"Proceedings of the 7th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP","first-page":"278","article-title":"Gemma scope: open sparse autoencoders everywhere all at once on gemma 2","author":"Lieberum","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib0920","first-page":"87","article-title":"Awq: activation-aware weight quantization for on-device LLM compression and acceleration","volume":"6","author":"Lin","year":"2024","journal-title":"Proc. Mach. Learn. Syst."},{"key":"10.1016\/j.cosrev.2026.101011_bib0925","author":"Lin"},{"key":"10.1016\/j.cosrev.2026.101011_bib0930","article-title":"On the biology of a large language model","author":"Lindsey","year":"2025","journal-title":"Transformer Circuits Thread"},{"key":"10.1016\/j.cosrev.2026.101011_bib0935","series-title":"Advances in Neural Information Processing Systems 36: Annual Conference on Neural Information Processing Systems 2023, NeurIPS 2023, New Orleans, LA, USA, December 10 - 16, 2023","article-title":"Visual instruction tuning","author":"Liu","year":"2023"},{"key":"10.1016\/j.cosrev.2026.101011_bib0940","author":"Liu"},{"key":"10.1016\/j.cosrev.2026.101011_bib0945","series-title":"Findings of the Association for Computational Linguistics, ACL 2025, Vienna, Austria, July 27 - August 1, 2025","first-page":"19243","article-title":"Sens-merging: sensitivity-guided parameter balancing for merging large language models","author":"Liu","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0950","article-title":"Routers in vision mixture of experts: an empirical study","volume":"2024","author":"Liu","year":"2024","journal-title":"Trans. Mach. Learn. Res."},{"key":"10.1016\/j.cosrev.2026.101011_bib0955","author":"Liu"},{"key":"10.1016\/j.cosrev.2026.101011_bib0960","series-title":"Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers)","first-page":"61","article-title":"P-Tuning: prompt tuning can be comparable to fine-tuning across scales and tasks","author":"Liu","year":"2022"},{"key":"10.1016\/j.cosrev.2026.101011_bib0965","series-title":"Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing","first-page":"992","article-title":"On relation-specific neurons in large language models","author":"Liu","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0970","author":"Liu"},{"key":"10.1016\/j.cosrev.2026.101011_bib0975","series-title":"Findings of the Association for Computational Linguistics: EMNLP 2025","first-page":"2121","article-title":"Tracing multilingual factual knowledge acquisition in pretraining","author":"Liu","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib0980","series-title":"The Eleventh International Conference on Learning Representations","article-title":"Omnigrok: grokking beyond algorithmic data","author":"Liu","year":"2023"},{"key":"10.1016\/j.cosrev.2026.101011_bib0985","author":"L\u00f3pez-Otal"},{"key":"10.1016\/j.cosrev.2026.101011_bib0990","author":"Lu"},{"key":"10.1016\/j.cosrev.2026.101011_bib0995","author":"Lu"},{"key":"10.1016\/j.cosrev.2026.101011_bib1000","author":"Lu"},{"key":"10.1016\/j.cosrev.2026.101011_bib1005","author":"Luo"},{"key":"10.1016\/j.cosrev.2026.101011_bib1010","series-title":"Forty-Second International Conference on Machine Learning","article-title":"Fast and low-cost genomic foundation models via outlier removal","author":"Luo","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1015","author":"Luo"},{"key":"10.1016\/j.cosrev.2026.101011_bib1020","author":"Lv"},{"key":"10.1016\/j.cosrev.2026.101011_bib1025","series-title":"The Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24\u201328, 2025","article-title":"Sparse feature circuits: discovering and editing interpretable causal graphs in language models","author":"Marks","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1030","series-title":"Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing","first-page":"7713","article-title":"Unlocking the future: exploring look-ahead planning mechanistic interpretability in large language models","author":"Men","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib1035","series-title":"Findings of the Association for Computational Linguistics: ACL 2025","first-page":"20192","article-title":"ShortGPT: layers in large language models are more redundant than you expect","author":"Men","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1040","series-title":"Advances in Neural Information Processing Systems 35: Annual Conference on Neural Information Processing Systems 2022, NeurIPS 2022, New Orleans, LA, USA, November 28 - December 9, 2022","article-title":"Locating and editing factual associations in GPT","author":"Meng","year":"2022"},{"key":"10.1016\/j.cosrev.2026.101011_bib1045","series-title":"The Eleventh International Conference on Learning Representations","article-title":"Mass-editing memory in a transformer","author":"Meng","year":"2023"},{"key":"10.1016\/j.cosrev.2026.101011_bib1050","author":"Miller"},{"key":"10.1016\/j.cosrev.2026.101011_bib1055","series-title":"ICLR 2025 Workshop on Building Trust in Language Models and Applications","article-title":"In-context meta learning induces multi-phase circuit emergence","author":"Minegishi","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1060","series-title":"The Sixth Workshop on Insights From Negative Results in NLP","first-page":"46","article-title":"Language-specific neurons do not facilitate cross-lingual transfer","author":"Mondal","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1065","series-title":"An Introduction to Psychology","author":"Morgan","year":"1927"},{"key":"10.1016\/j.cosrev.2026.101011_bib1070","series-title":"Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","first-page":"6326","article-title":"Exploring alignment in shared cross-lingual spaces","author":"Mousi","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib1075","author":"Mudide"},{"key":"10.1016\/j.cosrev.2026.101011_bib1080","series-title":"Forty-Second International Conference on Machine Learning, ICML 2025, Vancouver, BC, Canada, July 13\u201319, 2025","article-title":"MIB: a mechanistic interpretability benchmark","author":"Mueller","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1085","author":"Muhamed"},{"key":"10.1016\/j.cosrev.2026.101011_bib1090","series-title":"Findings of the Association for Computational Linguistics: NAACL 2025","first-page":"1604","article-title":"Decoding dark matter: specialized sparse autoencoders for interpreting rare concepts in foundation models","author":"Muhamed","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1095","series-title":"ICML 2025 Workshop on Reliable and Responsible Foundation Models","article-title":"The geometry of forgetting: analyzing machine unlearning through local learning coefficients","author":"Muhamed","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1100","author":"Nanda"},{"key":"10.1016\/j.cosrev.2026.101011_bib1105","series-title":"The Eleventh International Conference on Learning Representations","article-title":"Progress measures for grokking via mechanistic interpretability","author":"Nanda","year":"2023"},{"key":"10.1016\/j.cosrev.2026.101011_bib1110","series-title":"The Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24\u201328, 2025","article-title":"Towards interpreting visual information processing in vision-language models","author":"Neo","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1115","author":"Nguyen"},{"key":"10.1016\/j.cosrev.2026.101011_bib1120","series-title":"Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","first-page":"20619","article-title":"Multi-attribute steering of language models via targeted intervention","author":"Nguyen","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1125","author":"Nguyen"},{"key":"10.1016\/j.cosrev.2026.101011_bib1130","author":"Nguyen"},{"key":"10.1016\/j.cosrev.2026.101011_bib1135","author":"Nguyen"},{"key":"10.1016\/j.cosrev.2026.101011_bib1140","series-title":"Findings of the Association for Computational Linguistics: EMNLP 2025","first-page":"690","article-title":"Mechanistic understanding and mitigation of language confusion in english-centric large language models","author":"Nie","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1145","author":"Nikankin"},{"key":"10.1016\/j.cosrev.2026.101011_bib1150","series-title":"The Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24\u201328, 2025","article-title":"Arithmetic without algorithms: language models solve math with a bag of heuristics","author":"Nikankin","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1155","series-title":"Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","first-page":"16218","article-title":"LLaMA see, LLaMA do: a mechanistic perspective on contextual entrainment and distraction in LLMs","author":"Niu","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1165","author":"Notsawo Jr"},{"key":"10.1016\/j.cosrev.2026.101011_bib1170","series-title":"Findings of the Association for Computational Linguistics: EMNLP 2025, Suzhou, China, November 4\u20139, 2025","first-page":"18263","article-title":"Probing semantic routing in large mixture-of-expert models","author":"Olson","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1175","author":"Olsson"},{"key":"10.1016\/j.cosrev.2026.101011_bib1180","author":"Achiam"},{"key":"10.1016\/j.cosrev.2026.101011_bib1185","series-title":"The Thirteenth International Conference on Learning Representations","article-title":"LLMs know more than they show: on the intrinsic representation of LLM hallucinations","author":"Orgad","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1190","series-title":"The Thirty-Ninth Annual Conference on Neural Information Processing Systems","article-title":"Sparse autoencoders learn monosemantic features in vision-language models","author":"Pach","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1195","author":"Pai"},{"key":"10.1016\/j.cosrev.2026.101011_bib1200","series-title":"Forty-Second International Conference on Machine Learning","article-title":"The hidden dimensions of LLM alignment: a multi-dimensional analysis of orthogonal safety directions","author":"Pan","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1205","author":"Panigrahi"},{"key":"10.1016\/j.cosrev.2026.101011_bib1210","author":"Parrack"},{"key":"10.1016\/j.cosrev.2026.101011_bib1215","series-title":"Proceedings of the 5th Workshop on Research in Computational Linguistic Typology and Multilingual NLP","first-page":"22","article-title":"Identifying the correlation between language distance and cross-lingual transfer in a multilingual representation space","author":"Philippy","year":"2023"},{"key":"10.1016\/j.cosrev.2026.101011_bib1220","series-title":"Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)","first-page":"9661","article-title":"Beyond logit lens: contextual embeddings for robust hallucination detection & grounding in VLMs","author":"Phukan","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1225","series-title":"Findings of the Association for Computational Linguistics: ACL 2024","first-page":"11481","article-title":"Peering into the mind of language models: an approach for attribution in contextual question answering","author":"Phukan","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib1230","author":"Postmus"},{"key":"10.1016\/j.cosrev.2026.101011_bib1235","series-title":"Findings of the Association for Computational Linguistics: EMNLP 2025","first-page":"17735","article-title":"Can role vectors affect LLM behaviour?","author":"Potert\u00ec","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1240","author":"Prakash"},{"key":"10.1016\/j.cosrev.2026.101011_bib1245","doi-asserted-by":"crossref","DOI":"10.1016\/j.patter.2024.101118","article-title":"A survey of multilingual large language models","volume":"6","author":"Qin","year":"2025","journal-title":"Patterns"},{"key":"10.1016\/j.cosrev.2026.101011_bib1250","author":"Qiye"},{"key":"10.1016\/j.cosrev.2026.101011_bib1255","series-title":"The Twelfth International Conference on Learning Representations","article-title":"Understanding addition in transformers","author":"Quirke","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib1260","author":"Yang"},{"key":"10.1016\/j.cosrev.2026.101011_bib1265","first-page":"9","article-title":"Language models are unsupervised multitask learners","volume":"1","author":"Radford","year":"2019","journal-title":"OpenAI blog"},{"key":"10.1016\/j.cosrev.2026.101011_bib1270","series-title":"ICML 2025 Workshop on Methods and Opportunities at Small Scale","article-title":"The necessity for intervention fidelity: unintended side effects when steering LLMs","author":"Raedler","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1275","author":"Rai"},{"key":"10.1016\/j.cosrev.2026.101011_bib1280","author":"Raimondi"},{"key":"10.1016\/j.cosrev.2026.101011_bib1285","author":"Rajamanoharan"},{"key":"10.1016\/j.cosrev.2026.101011_bib1290","author":"Rajamanoharan"},{"key":"10.1016\/j.cosrev.2026.101011_bib1295","author":"Ranjan"},{"key":"10.1016\/j.cosrev.2026.101011_bib1300","series-title":"2023 IEEE Conference on Secure and Trustworthy Machine Learning (Satml)","first-page":"464","article-title":"Toward transparent AI: a survey on interpreting the inner structures of deep neural networks","author":"R\u00e4uker","year":"2023"},{"key":"10.1016\/j.cosrev.2026.101011_bib1305","series-title":"Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume, EACL 2021, Online, April 19 - 23, 2021","first-page":"3363","article-title":"Probing the probing paradigm: does probing accuracy entail task relevance?","author":"Ravichander","year":"2021"},{"key":"10.1016\/j.cosrev.2026.101011_bib1310","author":"Ravindran"},{"key":"10.1016\/j.cosrev.2026.101011_bib1315","series-title":"Proceedings of the 31st International Conference on Computational Linguistics","first-page":"2988","article-title":"Do large language models mirror cognitive language processing?","author":"Ren","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1320","author":"Ren"},{"key":"10.1016\/j.cosrev.2026.101011_bib1325","series-title":"Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing","first-page":"20465","article-title":"Explainability and interpretability of multilingual large language models: a survey","author":"Resck","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1330","series-title":"Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","first-page":"15504","article-title":"Steering LLaMA 2 via contrastive activation addition","author":"Rimsky","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib1335","author":"Sakarvadia"},{"key":"10.1016\/j.cosrev.2026.101011_bib1340","author":"Saphra"},{"key":"10.1016\/j.cosrev.2026.101011_bib1345","series-title":"Advances in Experimental Social Psychology","doi-asserted-by":"crossref","first-page":"1","DOI":"10.1016\/S0065-2601(08)60281-6","article-title":"Universals in the content and structure of values: theoretical advances and empirical tests in 20 countries","volume":"vol. 25","author":"Schwartz","year":"1992"},{"key":"10.1016\/j.cosrev.2026.101011_bib1350","author":"Sergeev"},{"key":"10.1016\/j.cosrev.2026.101011_bib1355","author":"Shang"},{"key":"10.1016\/j.cosrev.2026.101011_bib1360","series-title":"Proceedings of the IEEE\/CVF Conference on Computer Vision and Pattern Recognition","first-page":"11030","article-title":"Incremental residual concept bottleneck models","author":"Shang","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib1365","author":"Shao"},{"key":"10.1016\/j.cosrev.2026.101011_bib1370","author":"Sharma"},{"key":"10.1016\/j.cosrev.2026.101011_bib1375","author":"Shazeer"},{"key":"10.1016\/j.cosrev.2026.101011_bib1380","series-title":"5th International Conference on Learning Representations, ICLR 2017, Toulon, France, April 24\u201326, 2017, Conference Track Proceedings","article-title":"Outrageously large neural networks: the sparsely-gated mixture-of-experts layer","author":"Shazeer","year":"2017"},{"key":"10.1016\/j.cosrev.2026.101011_bib1385","series-title":"Proceedings of the 38th Pacific Asia Conference on Language, Information and Computation","first-page":"720","article-title":"Towards building efficient sentence BERT models using layer pruning","author":"Shelke","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib1390","series-title":"Advances in Neural Information Processing Systems 38: Annual Conference on Neural Information Processing Systems 2024, NeurIPS 2024, Vancouver, BC, Canada, December 10 - 15, 2024","article-title":"IRCAN: mitigating knowledge conflicts in LLM generation via identifying and reweighting context-aware neurons","author":"Shi","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib1395","series-title":"Proceedings of the 34th International Conference on Machine Learning (ICML)","article-title":"Learning important features through propagating activation differences","author":"Shrikumar","year":"2017"},{"key":"10.1016\/j.cosrev.2026.101011_bib1400","series-title":"Findings of the Association for Computational Linguistics: EMNLP 2025","first-page":"1690","article-title":"A survey on sparse autoencoders: interpreting the internal mechanisms of large language models","author":"Shu","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1405","series-title":"Proceedings of the 41st International Conference on Machine Learning","first-page":"45637","article-title":"What needs to go right for an induction head? A mechanistic study of in-context learning circuits and their formation","author":"Singh","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib1410","author":"Sinii"},{"key":"10.1016\/j.cosrev.2026.101011_bib1415","author":"Smilkov"},{"key":"10.1016\/j.cosrev.2026.101011_bib1420","series-title":"Proceedings of the 41st International Conference on Machine Learning","article-title":"Sparse is enough in fine-tuning pre-trained large language models","author":"Song","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib1425","author":"Soo"},{"key":"10.1016\/j.cosrev.2026.101011_bib1430","doi-asserted-by":"crossref","first-page":"79057","DOI":"10.52202\/079017-2510","article-title":"Vlg-cbm: training concept bottleneck models with vision-language guidance","volume":"37","author":"Srivastava","year":"2024","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"10.1016\/j.cosrev.2026.101011_bib1435","series-title":"Findings of the Association for Computational Linguistics: EMNLP 2024","first-page":"8189","article-title":"Activation scaling for steering and interpreting language models","author":"Stoehr","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib1440","series-title":"The Thirteenth International Conference on Learning Representations","article-title":"Improving instruction-following in language models through activation steering","author":"Stolfo","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1445","series-title":"Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing","first-page":"7035","article-title":"A mechanistic interpretation of arithmetic reasoning in language models using causal mediation analysis","author":"Stolfo","year":"2023"},{"key":"10.1016\/j.cosrev.2026.101011_bib1450","doi-asserted-by":"crossref","DOI":"10.1016\/j.neucom.2023.127063","article-title":"Roformer: enhanced transformer with rotary position embedding","volume":"568","author":"Su","year":"2024","journal-title":"Neurocomputing"},{"key":"10.1016\/j.cosrev.2026.101011_bib1455","series-title":"Findings of the Association for Computational Linguistics: EMNLP 2025","first-page":"9433","article-title":"Understanding how value neurons shape the generation of specified values in LLMs","author":"Su","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1460","author":"Su"},{"key":"10.1016\/j.cosrev.2026.101011_bib1465","author":"Su"},{"key":"10.1016\/j.cosrev.2026.101011_bib1470","author":"Su"},{"key":"10.1016\/j.cosrev.2026.101011_bib1475","series-title":"Forty-First International Conference on Machine Learning","article-title":"Whispering experts: neural interventions for toxicity mitigation in language models","author":"Suau","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib1480","series-title":"The Thirteenth International Conference on Learning Representations","article-title":"Concept bottleneck large language models","author":"Sun","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib1485","author":"Sun"},{"key":"10.1016\/j.cosrev.2026.101011_bib1490","series-title":"Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing","first-page":"24667","article-title":"Personality vector: modulating personality of large language models by model merging","author":"Sun","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1495","series-title":"Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing","first-page":"8122","article-title":"Probing for arithmetic errors in language models","author":"Sun","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1500","series-title":"Proceedings of the 34th International Conference on Machine Learning, ICML 2017, Sydney, NSW, Australia, 6\u201311 August 2017","first-page":"3319","article-title":"Axiomatic attribution for deep networks","author":"Sundararajan","year":"2017"},{"key":"10.1016\/j.cosrev.2026.101011_bib1505","series-title":"Proceedings of the 7th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP","first-page":"407","article-title":"Attribution patching outperforms automated circuit discovery","author":"Syed","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib1510","series-title":"Findings of the Association for Computational Linguistics: ACL 2025","first-page":"13090","article-title":"Mechanistic interpretability of emotion inference in large language models","author":"Tak","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1515","author":"Tan"},{"key":"10.1016\/j.cosrev.2026.101011_bib1520","series-title":"Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing","first-page":"6506","article-title":"Neuron specialization: leveraging intrinsic task modularity for multilingual machine translation","author":"Tan","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib1525","author":"Tan"},{"key":"10.1016\/j.cosrev.2026.101011_bib1530","series-title":"Pacific-Asia Conference on Knowledge Discovery and Data Mining","first-page":"56","article-title":"Interpreting pretrained language models via concept bottlenecks","author":"Tan","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib1535","author":"Tan"},{"key":"10.1016\/j.cosrev.2026.101011_bib1540","series-title":"The Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24\u201328, 2025","article-title":"RazorAttention: efficient KV cache compression through retrieval heads","author":"Tang","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1545","series-title":"Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","first-page":"5701","article-title":"Language-specific neurons: the key to multilingual capabilities in large language models","author":"Tang","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib1550","series-title":"Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","first-page":"6832","article-title":"Unlocking general long chain-of-thought reasoning capabilities of large language models via representation engineering","author":"Tang","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1555","author":"Tao"},{"key":"10.1016\/j.cosrev.2026.101011_bib1560","article-title":"Scaling monosemanticity: extracting interpretable features from claude 3 sonnet","author":"Templeton","year":"2024","journal-title":"Transformer Circuits Thread"},{"key":"10.1016\/j.cosrev.2026.101011_bib1565","series-title":"Proceedings of the 57th Conference of the Association for Computational Linguistics, ACL 2019, Florence, Italy, July 28- August 2, 2019, Volume 1: Long Papers","first-page":"4593","article-title":"BERT rediscovers the classical NLP pipeline","author":"Tenney","year":"2019"},{"key":"10.1016\/j.cosrev.2026.101011_bib1570","series-title":"Has It Trained yet? NeurIPS 2022 Workshop","article-title":"The slingshot mechanism: an empirical study of adaptive optimizers and the Grokking Phenomenon","author":"Thilak","year":"2022"},{"key":"10.1016\/j.cosrev.2026.101011_bib1575","series-title":"Findings of the Association for Computational Linguistics: EMNLP 2025","first-page":"18640","article-title":"Internal states before wait modulate reasoning patterns","author":"Troitskii","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1580","author":"Turner"},{"key":"10.1016\/j.cosrev.2026.101011_bib1585","author":"Valade"},{"key":"10.1016\/j.cosrev.2026.101011_bib1590","author":"Varma"},{"key":"10.1016\/j.cosrev.2026.101011_bib1595","author":"Venhoff"},{"key":"10.1016\/j.cosrev.2026.101011_bib1600","series-title":"Advances in Neural Information Processing Systems","first-page":"12388","article-title":"Investigating gender bias in language models using causal mediation analysis","author":"Vig","year":"2020"},{"key":"10.1016\/j.cosrev.2026.101011_bib1605","series-title":"Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics","first-page":"5797","article-title":"Analyzing multi-head self-attention: specialized heads do the heavy lifting, the rest can be pruned","author":"Voita","year":"2019"},{"key":"10.1016\/j.cosrev.2026.101011_bib1610","doi-asserted-by":"crossref","first-page":"95238","DOI":"10.52202\/079017-3017","article-title":"Grokking of implicit reasoning in transformers: a mechanistic journey to the edge of generalization","volume":"37","author":"Wang","year":"2024","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"10.1016\/j.cosrev.2026.101011_bib1615","series-title":"Findings of the Association for Computational Linguistics: EMNLP 2023","first-page":"4283","article-title":"HadSkip: homotopic and adaptive layer skipping of pre-trained language models for efficient inference","author":"Wang","year":"2023"},{"key":"10.1016\/j.cosrev.2026.101011_bib1620","author":"Wang"},{"key":"10.1016\/j.cosrev.2026.101011_bib1625","series-title":"The Eleventh International Conference on Learning Representations","article-title":"Interpretability in the wild: a circuit for indirect object identification in GPT-2 small","author":"Wang","year":"2023"},{"key":"10.1016\/j.cosrev.2026.101011_bib1630","series-title":"Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","first-page":"5075","article-title":"Lost in multilinguality: dissecting cross-lingual factual inconsistency in transformer language models","author":"Wang","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1635","series-title":"The Thirty-Ninth Annual Conference on Neural Information Processing Systems","article-title":"Two experts are all you need for steering thinking: reinforcing cognitive effort in MoE reasoning models without additional training","author":"Wang","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1640","series-title":"Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing","first-page":"2637","article-title":"Language mixing in reasoning language models: patterns, impact, and internal causes","author":"Wang","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1645","author":"Wang"},{"key":"10.1016\/j.cosrev.2026.101011_bib1650","author":"Wang"},{"key":"10.1016\/j.cosrev.2026.101011_bib1655","author":"Wang"},{"key":"10.1016\/j.cosrev.2026.101011_bib1660","series-title":"AAAI-25, Sponsored by the Association for the Advancement of Artificial Intelligence, February 25 - March 4, 2025, Philadelphia, PA, USA","first-page":"14432","article-title":"BrainMAP: learning multiple activation pathways in brain networks","author":"Wang","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1665","series-title":"The Thirteenth International Conference on Learning Representations","article-title":"Surgical, cheap, and flexible: mitigating false refusal in language models via single vector ablation","author":"Wang","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1670","series-title":"The Thirty-Ninth Annual Conference on Neural Information Processing Systems","article-title":"Refusal direction is universal across safety-aligned languages","author":"Wang","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1675","author":"Wang"},{"key":"10.1016\/j.cosrev.2026.101011_bib1680","author":"Wang"},{"key":"10.1016\/j.cosrev.2026.101011_bib1685","author":"Wang"},{"key":"10.1016\/j.cosrev.2026.101011_bib1690","author":"Wang"},{"key":"10.1016\/j.cosrev.2026.101011_bib1695","author":"Ward"},{"key":"10.1016\/j.cosrev.2026.101011_bib1700","author":"van der Weij"},{"key":"10.1016\/j.cosrev.2026.101011_bib1705","series-title":"Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","first-page":"15366","article-title":"Do LLaMAs work in english? On the latent language of multilingual transformers","author":"Wendler","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib1710","author":"Weng"},{"key":"10.1016\/j.cosrev.2026.101011_bib1715","author":"Wu"},{"key":"10.1016\/j.cosrev.2026.101011_bib1720","author":"Wu"},{"key":"10.1016\/j.cosrev.2026.101011_bib1725","series-title":"Forty-Second International Conference on Machine Learning","article-title":"AxBench: steering LLMs? Even simple baselines outperform sparse autoencoders","author":"Wu","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1730","series-title":"Advances in Neural Information Processing Systems 38: Annual Conference on Neural Information Processing Systems 2024, NeurIPS 2024, Vancouver, BC, Canada, December 10 - 15, 2024","article-title":"ReFT: representation finetuning for language models","author":"Wu","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib1735","series-title":"The Thirteenth International Conference on Learning Representations","article-title":"The semantic hub hypothesis: language models share semantic representations across languages and modalities","author":"Wu","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1740","author":"Wulff"},{"key":"10.1016\/j.cosrev.2026.101011_bib1745","series-title":"Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, EMNLP 2022, Abu Dhabi, United Arab Emirates, December 7\u201311, 2022","first-page":"8318","article-title":"Efficient adversarial training with robust early-bird tickets","author":"Xi","year":"2022"},{"key":"10.1016\/j.cosrev.2026.101011_bib1750","series-title":"Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing","first-page":"3351","article-title":"TokenSkip: controllable chain-of-thought compression in LLMs","author":"Xia","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1755","series-title":"International Conference on Machine Learning","first-page":"38087","article-title":"Smoothquant: accurate and efficient post-training quantization for large language models","author":"Xiao","year":"2023"},{"key":"10.1016\/j.cosrev.2026.101011_bib1760","author":"Xiao"},{"key":"10.1016\/j.cosrev.2026.101011_bib1765","unstructured":"G. Xiao, Y. Tian, B. Chen, S. Han, M. Lewis, Efficient streaming language models with attention sinks, arXiv (2023)."},{"key":"10.1016\/j.cosrev.2026.101011_bib1770","series-title":"Second Conference on Language Modeling","article-title":"Task-circuit quantization: leveraging knowledge localization and interpretability for compression","author":"Xiao","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1775","author":"Xiao"},{"key":"10.1016\/j.cosrev.2026.101011_bib1780","series-title":"International Conference on Learning Representations (ICLR)","article-title":"SORRY-bench: systematically evaluating large language model safety refusal","author":"Xie","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1785","series-title":"Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers)","first-page":"5725","article-title":"Importance-based neuron allocation for multilingual neural machine translation","author":"Xie","year":"2021"},{"key":"10.1016\/j.cosrev.2026.101011_bib1790","author":"Xiong"},{"key":"10.1016\/j.cosrev.2026.101011_bib1795","author":"Xiong"},{"key":"10.1016\/j.cosrev.2026.101011_bib1800","author":"Xiong"},{"key":"10.1016\/j.cosrev.2026.101011_bib1805","author":"Xiong"},{"key":"10.1016\/j.cosrev.2026.101011_bib1810","author":"Xiong"},{"key":"10.1016\/j.cosrev.2026.101011_bib1815","author":"Xiong"},{"key":"10.1016\/j.cosrev.2026.101011_bib1820","series-title":"Proceedings of the 31st International Conference on Computational Linguistics","first-page":"9393","article-title":"Let\u2019s focus on neuron: neuron-level supervised fine-tuning for large language model","author":"Xu","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1825","author":"Xu"},{"key":"10.1016\/j.cosrev.2026.101011_bib1830","author":"Xu"},{"key":"10.1016\/j.cosrev.2026.101011_bib1835","doi-asserted-by":"crossref","first-page":"7093","DOI":"10.52202\/075280-0310","article-title":"Ties-merging: resolving interference when merging models","volume":"36","author":"Yadav","year":"2023","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"10.1016\/j.cosrev.2026.101011_bib1840","series-title":"Advances in Neural Information Processing Systems 36: Annual Conference on Neural Information Processing Systems 2023, NeurIPS 2023, New Orleans, LA, USA, December 10 - 16, 2023","article-title":"TIES-merging: resolving interference when merging models","author":"Yadav","year":"2023"},{"key":"10.1016\/j.cosrev.2026.101011_bib1845","author":"Yan"},{"key":"10.1016\/j.cosrev.2026.101011_bib1850","author":"Yang"},{"key":"10.1016\/j.cosrev.2026.101011_bib1855","author":"Yang"},{"key":"10.1016\/j.cosrev.2026.101011_bib1860","author":"Yang"},{"key":"10.1016\/j.cosrev.2026.101011_bib1865","author":"Yang"},{"key":"10.1016\/j.cosrev.2026.101011_bib1870","author":"Yao"},{"key":"10.1016\/j.cosrev.2026.101011_bib1875","series-title":"Advances in Neural Information Processing Systems","first-page":"118571","article-title":"Knowledge circuits in pretrained transformers","author":"Yao","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib1880","doi-asserted-by":"crossref","first-page":"118571","DOI":"10.52202\/079017-3765","article-title":"Knowledge circuits in pretrained transformers","volume":"37","author":"Yao","year":"2024","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"10.1016\/j.cosrev.2026.101011_bib1885","series-title":"The Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24\u201328, 2025","article-title":"Physics of language models: Part 2.1, grade-school math and the hidden reasoning process","author":"Ye","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1890","series-title":"Proceedings of the AAAI Conference on Artificial Intelligence","first-page":"22128","article-title":"Fit and prune: fast and training-free visual token pruning for multi-modal large language models","author":"Ye","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1895","series-title":"Findings of the Association for Computational Linguistics: EMNLP 2025","first-page":"6377","article-title":"Understanding refusal in language models with sparse autoencoders","author":"Yeo","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1900","series-title":"Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing","first-page":"10436","article-title":"Towards faithful natural language explanations: a study using activation patching in large language models","author":"Yeo","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1905","doi-asserted-by":"crossref","first-page":"9474","DOI":"10.52202\/079017-0301","article-title":"Lofit: localized fine-tuning on LLM representations","volume":"37","author":"Yin","year":"2024","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"10.1016\/j.cosrev.2026.101011_bib1910","series-title":"Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing","first-page":"184","article-title":"Interpreting language models with contrastive explanations","author":"Yin","year":"2022"},{"key":"10.1016\/j.cosrev.2026.101011_bib1915","author":"Yin"},{"key":"10.1016\/j.cosrev.2026.101011_bib1920","series-title":"Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing","first-page":"7517","article-title":"Probabilistic soundness guarantees in LLM reasoning chains","author":"You","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1925","author":"Yu"},{"key":"10.1016\/j.cosrev.2026.101011_bib1930","author":"Yu"},{"key":"10.1016\/j.cosrev.2026.101011_bib1935","series-title":"Findings of the Association for Computational Linguistics: ACL 2025","first-page":"6092","article-title":"Mitigate position bias in LLMs via scaling a single hidden states channel","author":"Yu","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1940","series-title":"Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","first-page":"24914","article-title":"Chain-of-reasoning: towards unified mathematical reasoning in large language models via a multi-paradigm perspective","author":"Yu","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1945","series-title":"Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, EMNLP 2024, Miami, FL, USA, November 12\u201316, 2024","first-page":"3281","article-title":"How do large language models learn in-context? Query and key matrices of in-context heads are two towers for metric learning","author":"Yu","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib1950","series-title":"Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing","first-page":"3293","article-title":"Interpreting arithmetic mechanism in large language models through comparative neuron analysis","author":"Yu","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib1955","series-title":"Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, EMNLP 2024, Miami, FL, USA, November 12\u201316, 2024","first-page":"3267","article-title":"Neuron-level knowledge attribution in large language models","author":"Yu","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib1960","author":"Yu"},{"key":"10.1016\/j.cosrev.2026.101011_bib1965","series-title":"Findings of the Association for Computational Linguistics: EMNLP 2025","first-page":"7065","article-title":"Locate-then-merge: neuron-level parameter fusion for mitigating catastrophic forgetting in multimodal LLMs","author":"Yu","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1970","author":"Yu"},{"key":"10.1016\/j.cosrev.2026.101011_bib1975","series-title":"Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing","first-page":"11268","article-title":"Back attention: understanding and enhancing multi-hop reasoning in large language models","author":"Yu","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1980","series-title":"Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","first-page":"23078","article-title":"Native sparse attention: hardware-aligned and natively trainable sparse attention","author":"Yuan","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1985","author":"Yuan"},{"key":"10.1016\/j.cosrev.2026.101011_bib1990","series-title":"The First Workshop on the Interplay of Model Behavior and Model Internals","article-title":"Interpreting the latent structure of operator precedence in language models","author":"Yugeswardeenoo","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib1995","author":"Zeng"},{"key":"10.1016\/j.cosrev.2026.101011_bib2000","author":"Zhang"},{"key":"10.1016\/j.cosrev.2026.101011_bib2005","author":"Zhang"},{"key":"10.1016\/j.cosrev.2026.101011_bib2010","author":"Zhang"},{"key":"10.1016\/j.cosrev.2026.101011_bib2015","author":"Zhang"},{"key":"10.1016\/j.cosrev.2026.101011_bib2020","first-page":"1","article-title":"A question-centric multi-experts contrastive learning framework for improving the accuracy and interpretability of deep sequential knowledge tracing models","volume":"19","author":"Zhang","year":"2025","journal-title":"ACM Trans. Knowl. Discov. Data"},{"key":"10.1016\/j.cosrev.2026.101011_bib2025","series-title":"Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","first-page":"4818","article-title":"ShifCon: enhancing non-dominant language capabilities with a shift-based multilingual contrastive framework","author":"Zhang","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib2030","series-title":"Findings of the Association for Computational Linguistics, ACL 2024, Bangkok, Thailand and Virtual Meeting, August 11\u201316, 2024","first-page":"7467","article-title":"Balancing speciality and versatility: a coarse to fine framework for supervised fine-tuning large language model","author":"Zhang","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib2035","author":"Zhang"},{"key":"10.1016\/j.cosrev.2026.101011_bib2040","author":"Zhang"},{"key":"10.1016\/j.cosrev.2026.101011_bib2045","author":"Zhang"},{"key":"10.1016\/j.cosrev.2026.101011_bib2050","author":"Zhang"},{"key":"10.1016\/j.cosrev.2026.101011_bib2055","author":"Zhang"},{"key":"10.1016\/j.cosrev.2026.101011_bib2060","series-title":"Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","first-page":"8908","article-title":"TruthX: alleviating hallucinations by editing large language models in truthful space","author":"Zhang","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib2065","series-title":"Forty-First International Conference on Machine Learning","article-title":"Interpreting and improving large language models in arithmetic calculation","author":"Zhang","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib2070","series-title":"Proceedings of the 31st International Conference on Computational Linguistics, COLING 2025, Abu Dhabi, UAE, January 19\u201324, 2025","first-page":"5775","article-title":"Multilingual knowledge editing with language-agnostic factual neurons","author":"Zhang","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib2075","author":"Zhang"},{"key":"10.1016\/j.cosrev.2026.101011_bib2080","series-title":"Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","first-page":"1701","article-title":"Fine-tuning happens in tiny subspaces: exploring intrinsic task-specific subspaces of pre-trained language models","author":"Zhang","year":"2023"},{"key":"10.1016\/j.cosrev.2026.101011_bib2085","author":"Zhang"},{"key":"10.1016\/j.cosrev.2026.101011_bib2090","series-title":"IEEE\/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2025, Nashville, TN, USA, June 11\u201315, 2025","first-page":"19781","article-title":"Cross-modal information flow in multimodal large language models","author":"Zhang","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib2095","author":"Zhang"},{"key":"10.1016\/j.cosrev.2026.101011_bib2100","series-title":"Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), ACL 2024, Bangkok, Thailand, August 11\u201316, 2024","first-page":"6228","article-title":"Unveiling linguistic regions in large language models","author":"Zhang","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib2105","author":"Zhao"},{"key":"10.1016\/j.cosrev.2026.101011_bib2110","series-title":"Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing","first-page":"28518","article-title":"Probing and boosting large language models capabilities via attention heads","author":"Zhao","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib2115","first-page":"1","article-title":"Explainability for large language models: a survey","volume":"15","author":"Zhao","year":"2024","journal-title":"ACM Trans. Intell. Syst. Technol."},{"key":"10.1016\/j.cosrev.2026.101011_bib2120","author":"Zhao"},{"key":"10.1016\/j.cosrev.2026.101011_bib2125","series-title":"The Thirty-Ninth Annual Conference on Neural Information Processing Systems","article-title":"LLMs encode harmfulness and refusal separately","author":"Zhao","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib2130","series-title":"Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)","first-page":"5117","article-title":"Steering knowledge selection behaviours in LLMs via SAE-based representation engineering","author":"Zhao","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib2135","author":"Zhao"},{"key":"10.1016\/j.cosrev.2026.101011_bib2140","series-title":"Advances in Neural Information Processing Systems","first-page":"15296","article-title":"How do large language models handle multilingualism?","author":"Zhao","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib2145","series-title":"The Thirteenth International Conference on Learning Representations","article-title":"Understanding and enhancing safety mechanisms of LLMs via safety-specific neuron","author":"Zhao","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib2150","author":"Zheng"},{"key":"10.1016\/j.cosrev.2026.101011_bib2155","series-title":"The Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7\u201311, 2024","article-title":"Seeking neural nuggets: knowledge transfer in large language models from a parametric perspective","author":"Zhong","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib2160","series-title":"Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation, LREC\/COLING 2024, 20\u201325 May, 2024, Torino, Italy","first-page":"12527","article-title":"ORTicket: let one robust BERT ticket transfer across different tasks","author":"Zhou","year":"2024"},{"key":"10.1016\/j.cosrev.2026.101011_bib2165","series-title":"Advances in Neural Information Processing Systems 35: Annual Conference on Neural Information Processing Systems 2022, NeurIPS 2022, New Orleans, LA, USA, November 28 - December 9, 2022","article-title":"Mixture-of-experts with expert choice routing","author":"Zhou","year":"2022"},{"key":"10.1016\/j.cosrev.2026.101011_bib2170","author":"Zhou"},{"key":"10.1016\/j.cosrev.2026.101011_bib2175","series-title":"The Thirteenth International Conference on Learning Representations","article-title":"On the role of attention heads in large language model safety","author":"Zhou","year":"2025"},{"key":"10.1016\/j.cosrev.2026.101011_bib2180","series-title":"Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","first-page":"12135","article-title":"LANDeRMT: dectecting and routing language-aware neurons for selectively finetuning LLMs to machine translation","author":"Zhu","year":"2024"}],"container-title":["Computer Science Review"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/api.elsevier.com\/content\/article\/PII:S157401372600119X?httpAccept=text\/xml","content-type":"text\/xml","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/api.elsevier.com\/content\/article\/PII:S157401372600119X?httpAccept=text\/plain","content-type":"text\/plain","content-version":"vor","intended-application":"text-mining"}],"deposited":{"date-parts":[[2026,7,6]],"date-time":"2026-07-06T18:02:38Z","timestamp":1783360958000},"score":1,"resource":{"primary":{"URL":"https:\/\/linkinghub.elsevier.com\/retrieve\/pii\/S157401372600119X"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,11]]},"references-count":435,"alternative-id":["S157401372600119X"],"URL":"https:\/\/doi.org\/10.1016\/j.cosrev.2026.101011","relation":{},"ISSN":["1574-0137"],"issn-type":[{"value":"1574-0137","type":"print"}],"subject":[],"published":{"date-parts":[[2026,11]]},"assertion":[{"value":"Elsevier","name":"publisher","label":"This article is maintained by"},{"value":"Locate, steer, and improve: A practical survey of actionable mechanistic interpretability in large language models","name":"articletitle","label":"Article Title"},{"value":"Computer Science Review","name":"journaltitle","label":"Journal Title"},{"value":"https:\/\/doi.org\/10.1016\/j.cosrev.2026.101011","name":"articlelink","label":"CrossRef DOI link to publisher maintained version"},{"value":"article","name":"content_type","label":"Content Type"},{"value":"\u00a9 2026 Elsevier Inc. All rights are reserved, including those for text and data mining, AI training, and similar technologies.","name":"copyright","label":"Copyright"}],"article-number":"101011"}}