REVIEW 2 major objections 4 minor 51 references
Deployable Per-Instance Multi-Layer Activation Steering for Large Language Models
T0 review · 2 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash
Pith's one-line read The best layers for activation steering are a per-input decision, and a label-free system can learn to pick them.
desk verdict Solid empirical paper on per-instance layer steering; the K=3 results hold up, and the K>3 caveat is honestly flagged. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The argument is carried by a few named components. The per-instance oracle scores every layer subset against the gold answer and supplies the ceiling of what selection can achieve. The Top-K-Marginal rule (TKM) ranks layers by their single-layer effect $m_l(x)$ and picks the top $K$; at linear cost in depth it nearly attains the exhaustive optimum, which is why it becomes the training target rather than a deployable system. A game-theoretic credit allocation over the chosen subset shows that the top-marginal layer is also the joint subset's top credit-bearer, explaining the greedy match. W2S-Multi is a single-hidden-layer MLP over a 25-dimensional PCA of the prompt embedding, trained with a listwise softmax-KL loss to reproduce each input's per-layer effect distribution. A logistic direction classifier on the same features infers which answer the steering should move toward, and an adaptive-$K$ gate walks the ranked prefix, scores short steered passes against that inferred direction, and stops on a lift plateau, a Yes/No mass floor, or a mass drop, so it steers no more layers than necessary.
What would settle it
Run the same protocol on a transparent model with few enough layers (e.g., 16) that exhaustive subset search is feasible at $K=4$ and $K=5$, then compare TKM's lift against Exhaustive and W2S-Multi's recovered lift against both; if TKM falls materially below Exhaustive, or the deployable ranker's recovery drops toward or below the fixed global baseline at those $K$, the central claim that per-instance selection is learnable and deployable would be refuted.
Extended reading notes
Core claim
The central claim is that injection layers are an instance-level decision rather than a hyperparameter, and that the per-instance optimum is learnable from the prompt alone. A per-instance oracle over subsets of the 32 layers shows the best layers vary from one input to the next, and the greedy Top-K-Marginal rule—taking the top $K$ layers by single-layer lift $m_l(x)$—is statistically indistinguishable from the exhaustive $K=3$ optimum on 15 of the 24 full-set configurations and 16 of 24 on the steerable stratum. The match is structural: the effective steering vectors sit in a narrow, nearly collinear mid-band, neighbouring layers behave as near-substitutes, and the exhaustive optimum often pads its picks with near-zero-effect bottom layers, so the top-marginal layer is almost always the joint subset's top credit-bearer. The deployable system W2S-Multi reads no gold label at inference and recovers 93% of the exhaustive oracle's steerable lift on Llama-3-8B-Instruct and 65% on Aya-Expanse-8B, while staying at or above the unsteered baseline on average and largely avoiding the fluency collapse that fixed high-$K$ global sets incur.
Load-bearing premise
The load-bearing premise is that the greedy Top-K-Marginal rule faithfully approximates the exhaustive per-instance optimum, because the deployable ranker is trained to reproduce TKM's ranking; this equivalence has been verified only at $K \le 3$, while the $K=4,5$ trend rests on a beam heuristic, so if the greedy proxy degrades at higher $K$, on other models, or on other tasks, the ranker's training target is wrong and the deployed subset will not recover the oracle's lift.
Editorial extensions
If this is right
- Fixed global layer sets are not a safe default: they corrupt already-correct inputs on the saturated stratum, while per-instance selection captures the upside without sacrificing baseline alignment.
- Because the ranker, direction classifier, and gate are label-free, per-instance multi-layer steering can run at deployment with one embedding pass and at most a few short steered passes.
- Fluency collapse is a consequence of fixed aggressive dose rather than of any particular layer picker; an adaptive gate that stops on a lift plateau removes most of the collapse at a small behavioural cost.
- The direction-over-magnitude account predicts that a misdirected global set erodes one answer class while helping the other, and that saturated unflippable inputs are a ceiling of static-coefficient steering, not a selector defect.
- Practitioners should treat layer choice as part of per-input inference-time control, not as a hyperparameter fixed once per task.
Reading between the lines
- A testable extension the paper leaves open: on a model with few enough layers for exhaustive search at $K=4,5$, TKM's near-equivalence to the oracle can be checked directly rather than through the beam heuristic, and the ranker's recovery can be measured against a true higher-$K$ ceiling.
- The 93% versus 65% recovery gap may be driven in large part by saturation share (80% of Aya-Expanse-8B test inputs are already saturated against 42% for Llama-3-8B-Instruct), which would mean the deployable recipe's headroom is model-dependent in a predictable way.
- The direction classifier's high AUC suggests it could be reused as a standalone abstention or refusal controller, since it already predicts the sign of the target answer from the prompt embedding.
- The paper's own limitations list notes that evaluation is in-distribution on yes/no persona questions; a natural next step is open-ended generation with human preference, where the $\Delta p$ metric may not capture selection benefits.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper argues that the choice of activation-steering layers is an instance-level decision, not a fixed global configuration, and proposes a fully deployable, label-free pipeline for per-instance multi-layer selection. On two 8B open-weight models and six binary persona traits, it first establishes an exhaustive per-instance oracle at K≤3 and shows that a greedy Top-K-Marginal (TKM) rule nearly matches it; it then trains a prompt-embedding ranker (W2S-Multi) to reproduce TKM's per-layer marginal distribution, adds a logistic direction classifier to infer the steering sign, and wraps both in an adaptive-K gate that walks the ranked prefix and stops on plateau, Y/N-mass floor, or backoff conditions. The paper reports that the deployed recipe recovers 93% of the oracle's steerable-stratum lift on Llama-3-8B-Instruct and 65% on Aya-Expanse-8B, never drives a trait-model cell below its unsteered baseline on average, and largely avoids the fluency collapse of aggressive global selection, with a mechanistic 'direction over magnitude' account proposed to unify the failure modes.
Significance. If the results hold, this is a genuinely useful advance: it turns per-instance layer selection from a gold-scored oracle into a practical inference-time system, with strong empirical controls. The paper deserves credit for the instance-transfer permutation test (p_BH ≤ 0.011 in all twelve cells), the Shapley-closure analysis (91–100% full-set closure), the paired Wilcoxon/BH-FDR testing across configurations, the sensitivity sweeps of the steerable cutoff and gate constants, and the public code release. The aggregate gate result, which moves W2S-Multi from 8.6 to 8.8 pp while cutting perplexity cost, is a concrete and falsifiable demonstration of the deployability claim. The main weakness is that the central proxy (TKM as a surrogate for the exhaustive optimum) is verified only at K≤3, while the deployed gate operates at K=4–5 on a substantial fraction of instances.
major comments (2)
- [§3, §4.3, App. D, Limitations] The load-bearing link in the chain Exhaustive → TKM → W2S-Multi → adaptive-K gate is the claim that TKM faithfully approximates the per-instance optimum. That link is verified only at K≤3, where C(32,3)=4,960 subsets are enumerable; for K=4 and K=5, the paper's own ceiling is the pooled Beam heuristic (§3), and the Limitations explicitly state that the tractability result is verified only at K≤3. Because Fig. 4 shows the deployed gate putting substantial per-instance mass at K=4–5 on Llama-3-8B-Instruct (mean depth 2.58, with mass spread over every depth), and because W2S-Multi is trained to reproduce TKM's per-layer marginals (App. B), the §4.3 claim that W2S-Multi 'tracks the TKM ceiling at every dose' (Fig. 5) is not directly supported above K=3. I ask for a restricted exhaustive or pruned-enumeration check at K=4, or, failing that, a rephrasing that limits 'tracks the ceiling' to K≤3 and explicitly labels the K=4–5 behavior as indicative. The aggregate gate result (8.8 pp vs 8.6 pp at fixed K=3) bounds the practical harm, but the structural claim is currently stronger than the evidence.
- [§4.1 / App. D, Table 17] The statement that the exhaustive-minus-TKM gap is 'at or below about 2.6 percentage points in every cell' is presented as the headline evidence for TKM matching the oracle. That bound is for the full test set; the steerable-stratum column of Table 17 shows materially larger gaps on Aya-Expanse-8B (8.2 pp on CONSCIENTIOUSNESS, 4.7 pp on PHENOMENAL CONSCIOUSNESS), and Fig. 8 reports TKM recovery as low as 68% on that model. Since the paper's central recovery numbers (93%/65%) are computed on the steerable stratum, the full-set bound should not stand in for the steerable result in the text. Please state the steerable gaps alongside the full-set bound and discuss how the 65% W2S-Multi recovery on Aya-Expanse-8B relates to this weaker TKM agreement on the same stratum.
minor comments (4)
- [§3 / Fig. 3 caption] The body text says Beam is reported only at K∈{4,5}, and the Fig. 3 caption says Beam is 'defined only at K≥4', but Table 25 reports Beam at K=3. Please reconcile the definition or the table.
- [Fig. 3] The caption says 'Stars mark each method's adaptive gate', but the markers in the figure are not labelled in a legend; please add explicit legend entries so the gated variants are identifiable in grayscale.
- [§4.2 / App. F] The mechanistic trace in Fig. 21 is a single sampled instance and the 'catastrophic band' in Fig. 22 is described as qualitative; this is fine as illustration, but the text should state more prominently that the mechanistic account for collapse is based on one cell and one instance, not a population-level fit.
- [§4.3] The sentence 'the recipe recovers 93% ... and 65%' is immediately followed by a comparison with GT-global; it would help to also remind the reader that 80% of Aya-Expanse-8B test inputs are saturated and excluded from the steerable stratum, so the 65% applies to a small subset of that model's test inputs.
Circularity Check
No significant circularity: the gold-scored TKM target is an explicit supervised training signal, and the deployable system is evaluated on held-out test instances against an independently computed exhaustive oracle.
full rationale
The paper's derivation chain is an empirical pipeline, not a definitional circle. The per-instance oracle (Exhaustive) and the greedy proxy TKM are both gold-scored and are used only as upper bounds and as the training target for W2S-Multi; the paper says so explicitly: 'the rule instead becomes the target a prompt-only predictor is trained to reproduce' (Abstract) and 'TKM ... becomes the training target for what is' (§1). Training a predictor to imitate a supervised target and then measuring held-out recovery against the independent exhaustive oracle is standard supervised learning, not a fitted input renamed as a prediction: W2S-Multi is trained on a 200-prompt split and evaluated on a disjoint 100-prompt test split (Table 1, App. B), and its recovery is read against Exhaustive, not against its own training target. The load-bearing link TKM≈Exhaustive is verified empirically at K=3 against the enumerated optimum on 15–16 of 24 configurations, with the full-set gap ≤2.6 pp in every cell (App. D), and the per-instance character of the oracle is checked by a same-sign instance-transfer permutation control (App. Table 21). The paper contains no self-citations, and no load-bearing claim rests on a prior work by these authors. The limitation that K=4–5 trends rely on the Beam heuristic is an acknowledged scope restriction, not a circular reduction; the gated W2S-Multi's aggregate lift is additionally anchored at K=3 and compared against external deployable baselines (LN-global, All-layers). The adaptive-K gate's constants are set a priori with leave-one-cell-out and leave-one-task-out selection checks (App. C). No step in the derivation reduces by construction to its own inputs.
Assumptions & free parameters
free parameters (4)
- Gate constants (epsilon, phi, delta, Kmax_high, Kmax_low, base_prob_cutoff) =
0.001, 0.3, 0.30, 5, 3, 0.5
- PCA dimensionality for prompt embedding =
25
- Steerable stratum cutoff Pbase(g|x) =
0.99
- MLP hyperparameters (hidden, dropout, epochs, lr, weight_decay) =
64, 0.3, 300, 1e-3, 1e-4
assumptions (5)
- domain assumption CAA contrastive activation addition with last-token injection is a valid steering operator on frozen models.
- domain assumption The restricted Y/N softmax alignment lift (Delta p) is the correct intrinsic metric for persona steering.
- domain assumption Gold labels from Anthropic-Persona, as selected by Sun et al. (2025), are correct.
- ad hoc to paper Top-K-Marginal greedy ranking is a sufficient proxy for the exhaustive per-instance optimum.
- domain assumption Beam search (width 4) is a usable oracle approximation at K=4,5.
Cite this review
Pith. "Pith review of Deployable Per-Instance Multi-Layer Activation Steering for Large Language Models." pith.science (2026). https://pith.science/paper/M4BKYW6J
@misc{pith2026260808829,
author = {Pith},
title = {Pith review of: Deployable Per-Instance Multi-Layer Activation Steering for Large Language Models},
year = {2026},
howpublished = {\url{https://pith.science/paper/M4BKYW6J}},
note = {Machine review of arXiv:2608.08829}
}
read the original abstract
Activation steering edits the behaviour of a frozen language model by adding a learned vector to its residual stream, and current practice fixes the injection layers globally per task. We argue that the best layers are an instance-level decision, and we make per-instance, multi-layer selection both well understood and deployable. On two open-weight 8B models and six binary persona traits, a per-instance oracle over layer subsets shows that the best layers vary from one input to the next: on most trait-model pairs, no fixed global layer set recovers the per-instance benefit. A greedy rule that ranks layers by single-layer marginal effect recovers nearly all of the oracle's benefit, but both must score candidate layers against the gold answer, so neither can run at deployment; the rule instead becomes the target a prompt-only predictor is trained to reproduce. Our deployable recipe needs no label at inference: a per-instance layer ranker read off the prompt embedding, a classifier that infers the steering direction, and an adaptive gate that scores short steered passes against that inferred direction and steers no more layers than necessary. The recipe recovers most of the oracle's lift (the bulk on the stronger model, a clear majority on the harder one), never drives any trait-model pair below its unsteered alignment baseline on average, and largely avoids the fluency collapse that strong global selection incurs at higher layer counts. A mechanistic account, "direction over magnitude", explains the behavioural flip under a mis-directed global set, the output collapse from steering too many layers, and the ceiling of unsteerable inputs.
Figures
Figures from the paper (21 more)
Reference graph
Works this paper leans on
-
[1]
Pranav Bhandari, Nicolas Fay, Sanjeevan Selvaganapathy, Amitava Datta, Usman Naseem, and Mehwish Nasim. 2026. Activation-space personality steering: Hybrid layer selection for stable trait control in LLMs . In Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers), pages 6388--64...
arXiv 2026
-
[3]
Diaoul \'e Diallo, Katharina Dworatzyk, Sophie Jentzsch, Peer Sch \"u tt, Sabine Theis, and Tobias Hecking. 2025. The effectiveness of style vectors for steering large language models: A human evaluation. IEEE Access, 13:191443--191457. ArXiv:2601.21505; DOI 10.1109/ACCESS.2025.3628500
-
[4]
Soham Gadgil, Chris Lin, and Su-In Lee. 2026. Where to steer: Input-dependent layer selection for steering improves LLM alignment. arXiv preprint arXiv:2604.03867. Under review
arXiv 2026
-
[6]
Amoukou, Tom Bewley, Saumitra Mishra, and Manuela Veloso
Anna Hedstr \"o m, Salim I. Amoukou, Tom Bewley, Saumitra Mishra, and Manuela Veloso. 2025. To steer or not to steer? M echanistic error reduction with abstention for language models. In Proceedings of the 42nd International Conference on Machine Learning (ICML), pages 22924--22945. ArXiv:2510.13290
arXiv 2025
-
[7]
Kalervo J \"a rvelin and Jaana Kek \"a l \"a inen. 2002. Cumulated gain-based evaluation of IR techniques. ACM Transactions on Information Systems (TOIS), 20(4):422--446
work page 2002
-
[8]
Bruce W. Lee, Inkit Padhi, Karthikeyan Natesan Ramamurthy, Erik Miehling, Pierre Dognin, Manish Nagireddy, and Amit Dhurandhar. 2025. Programming refusal with conditional activation steering. In Proceedings of the International Conference on Learning Representations (ICLR). ArXiv:2409.05907
arXiv 2025
-
[9]
Kenneth Li, Oam Patel, Fernanda Vi \'e gas, Hanspeter Pfister, and Martin Wattenberg. 2023. Inference-time intervention: Eliciting truthful answers from a language model. Advances in Neural Information Processing Systems, 36. ArXiv:2306.03341
arXiv 2023
-
[10]
Duy Nguyen, Archiki Prasad, Elias Stengel-Eskin, and Mohit Bansal. 2025. Multi-attribute steering of language models via targeted intervention. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 20619--20634. ArXiv:2502.12446
arXiv 2025
Show all 51 references
-
[11]
Narmeen Fatimah Oozeer, Luke Marks, Fazl Barez, and Amir Abdullah. 2025. Beyond linear steering: Unified multi-attribute control for language models. In Findings of the Association for Computational Linguistics: EMNLP 2025, pages 23513--23557. Association for Computational Lin...
2025 arXiv
-
[12]
Jayneel Parekh, Pegah Khayatan, Mustafa Shukor, Arnaud Dapogny, Alasdair Newson, and Matthieu Cord. 2025. Learning to steer: Input-dependent steering for multimodal LLMs . In Advances in Neural Information Processing Systems (NeurIPS). ArXiv:2508.12815
2025
-
[13]
Ethan Perez, Sam Ringer, Kamil \.e Luko s i \=u t \.e , Karina Nguyen, Edwin Chen, Scott Heiner, Craig Pettit, Catherine Olsson, Sandipan Kundu, Saurav Kadavath, Andy Jones, Anna Chen, Ben Mann, Brian Israel, Bryan Seethor, Cameron McKinnon, Christopher Olah, Da Yan, Daniela A...
2023 arXiv
-
[14]
Nina Rimsky, Nick Gabrieli, Julian Schulz, Meg Tong, Evan Hubinger, and Alexander Turner. 2024. Steering Llama 2 via contrastive activation addition. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 15504...
2024 arXiv
-
[15]
Pau Rodriguez, Arno Blaas, Michal Klein, Luca Zappella, Nicholas Apostoloff, Marco Cuturi, and Xavier Suau. 2025. Controlling language and diffusion models by transporting activations. In Proceedings of the International Conference on Learning Representations (ICLR). ArXiv:2410.23054
2025 arXiv
-
[16]
Daniel Scalena, Gabriele Sarti, and Malvina Nissim. 2024. Multi-property steering of large language models with dynamic activation composition. In Proceedings of the 7th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP, pages 577--603. ArXiv:2406.17563
2024 arXiv
-
[17]
Lloyd S. Shapley. 1953. A value for n-person games. In H. W. Kuhn and A. W. Tucker, editors, Contributions to the Theory of Games II, volume 28 of Annals of Mathematics Studies, pages 307--317. Princeton University Press
1953
-
[18]
Samuel Soo, Wesley Teng, Balaganesh Chandrasekaran, Guoxian Tan, and Ming Yan. 2025. Interpretable steering of large language models with feature guided activation additions. In ICLR Workshop on Building Trust in LLMs and LLM Applications . ArXiv:2501.09929
2025 arXiv
-
[19]
Hao Sun, Huailiang Peng, Qiong Dai, Xu Bai, and Yanan Cao. 2025. LayerNavigator : Finding promising intervention layers for efficient activation steering in large language models. In Advances in Neural Information Processing Systems (NeurIPS). OpenReview wj4lM45xQR
2025
-
[20]
Daniel Tan, David Chanin, Aengus Lynch, Brooks Paige, Dimitrios Kanoulas, Adri \`a Garriga-Alonso, and Robert Kirk. 2024. Analysing the generalisation and reliability of steering vectors. In Advances in Neural Information Processing Systems (NeurIPS). ArXiv:2407.12404
2024 arXiv
-
[22]
Minh Hieu Vu and Tan M. Nguyen. 2025. Angular steering: Behavior control via rotation in activation space. In Advances in Neural Information Processing Systems (NeurIPS). Spotlight; arXiv:2510.26243
2025
-
[23]
Tianlong Wang, Xianfeng Jiao, Yinghao Zhu, Zhongzhi Chen, Yifan He, Xu Chu, Junyi Gao, Yasha Wang, and Liantao Ma. 2025. Adaptive activation steering: A tuning-free LLM truthfulness improvement method for diverse hallucinations categories. In Proceedings of the ACM Web Confere...
2025 arXiv
-
[24]
Lyucheng Wu, Mengru Wang, Ziwen Xu, Tri Cao, Nay Oo, Bryan Hooi, and Shumin Deng. 2025. Automating steering for safe multimodal large language models. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 792--814. ArXiv:2507.13255
2025
-
[25]
Yanzhao Zhang, Mingxin Li, Dingkun Long, Xin Zhang, Huan Lin, Baosong Yang, Pengjun Xie, An Yang, Dayiheng Liu, Junyang Lin, Fei Huang, and Jingren Zhou. 2025. Qwen3 Embedding : Advancing text embedding and reranking through foundation models. arXiv preprint arXiv:2506.05176
2025 arXiv
-
[26]
Andy Zou, Long Phan, Sarah Chen, James Campbell, Phillip Guo, Richard Ren, Alexander Pan, Xuwang Yin, Mantas Mazeika, Ann-Kathrin Dombrowski, Shashwat Goel, Nathaniel Li, Michael J Byun, Zifan Wang, Alex Mallen, Steven Basart, Sanmi Koyejo, Dawn Song, Matt Fredrikson, and 2 ot...
2023 arXiv
-
[27]
Amir Zur, Zhuofan Ying, Alexander Russell Loftus, Kerem S ahin, Steven Yu, Lucia Quirke, Tamar Rott Shaham, Natalie Shapira, Hadas Orgad, and David Bau. 2025. Token entanglement in subliminal learning. In Mechanistic Interpretability Workshop at NeurIPS 2025. OpenReview auKgpBRzIW
2025
-
[28]
Aaron Grattafiori and Abhimanyu Dubey and Abhinav Jauhri and Abhinav Pandey and Abhishek Kadian and Ahmad Al-Dahle and Aiesha Letman and Akhil Mathur and Alan Schelten and Alex Vaughan and Amy Yang and Angela Fan and Anirudh Goyal and Anthony Hartshorn and Aobo Yang and Archi ...
-
[29]
arXiv preprint arXiv:2412.04261 , year=
John Dang and Shivalika Singh and Daniel D'souza and Arash Ahmadian and Alejandro Salamanca and Madeline Smith and Aidan Peppin and Sungjin Hong and Manoj Govindassamy and Terrence Zhao and Sandra Kublik and Meor Amer and Viraat Aryabumi and Jon Ander Campos and Yi-Chern Tan a...
-
[30]
2025 , note=
Sun, Hao and Peng, Huailiang and Dai, Qiong and Bai, Xu and Cao, Yanan , booktitle=. 2025 , note=
2025
-
[31]
arXiv preprint arXiv:2308.10248 , year=
Steering Language Models With Activation Engineering , author=. arXiv preprint arXiv:2308.10248 , year=
-
[32]
Steering
Rimsky, Nina and Gabrieli, Nick and Schulz, Julian and Tong, Meg and Hubinger, Evan and Turner, Alexander , booktitle=. Steering. 2024 , note=
2024
-
[33]
Findings of the Association for Computational Linguistics: ACL 2023 , pages=
Discovering Language Model Behaviors with Model-Written Evaluations , author=. Findings of the Association for Computational Linguistics: ACL 2023 , pages=. 2023 , note=
2023
-
[34]
Where to Steer: Input-Dependent Layer Selection for Steering Improves
Gadgil, Soham and Lin, Chris and Lee, Su-In , journal=. Where to Steer: Input-Dependent Layer Selection for Steering Improves. 2026 , note=
2026
-
[35]
Proceedings of the International Conference on Learning Representations (ICLR) , year=
Programming Refusal with Conditional Activation Steering , author=. Proceedings of the International Conference on Learning Representations (ICLR) , year=
-
[36]
Activation-Space Personality Steering: Hybrid Layer Selection for Stable Trait Control in
Bhandari, Pranav and Fay, Nicolas and Selvaganapathy, Sanjeevan and Datta, Amitava and Naseem, Usman and Nasim, Mehwish , booktitle=. Activation-Space Personality Steering: Hybrid Layer Selection for Stable Trait Control in. 2026 , note=
2026
-
[37]
Advances in Neural Information Processing Systems , volume=
Inference-Time Intervention: Eliciting Truthful Answers from a Language Model , author=. Advances in Neural Information Processing Systems , volume=. 2023 , note=
2023
-
[38]
Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing (EMNLP) , pages=
Automating Steering for Safe Multimodal Large Language Models , author=. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing (EMNLP) , pages=. 2025 , note=
2025
-
[39]
Learning to Steer: Input-dependent Steering for Multimodal
Parekh, Jayneel and Khayatan, Pegah and Shukor, Mustafa and Dapogny, Arnaud and Newson, Alasdair and Cord, Matthieu , booktitle=. Learning to Steer: Input-dependent Steering for Multimodal. 2025 , note=
2025
-
[40]
To Steer or Not to Steer?
Hedstr. To Steer or Not to Steer?. Proceedings of the 42nd International Conference on Machine Learning (ICML) , pages=. 2025 , note=
2025
-
[41]
Proceedings of the International Conference on Learning Representations (ICLR) , year=
Controlling Language and Diffusion Models by Transporting Activations , author=. Proceedings of the International Conference on Learning Representations (ICLR) , year=
-
[42]
Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=
Multi-Attribute Steering of Language Models via Targeted Intervention , author=. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=. 2025 , note=
2025
-
[43]
Mechanistic Interpretability Workshop at NeurIPS 2025 , year=
Token Entanglement in Subliminal Learning , author=. Mechanistic Interpretability Workshop at NeurIPS 2025 , year=
2025
-
[44]
Zico and Hendrycks, Dan , journal=
Zou, Andy and Phan, Long and Chen, Sarah and Campbell, James and Guo, Phillip and Ren, Richard and Pan, Alexander and Yin, Xuwang and Mazeika, Mantas and Dombrowski, Ann-Kathrin and Goel, Shashwat and Li, Nathaniel and Byun, Michael J and Wang, Zifan and Mallen, Alex and Basar...
-
[45]
Advances in Neural Information Processing Systems (NeurIPS) , year=
Analysing the Generalisation and Reliability of Steering Vectors , author=. Advances in Neural Information Processing Systems (NeurIPS) , year=
-
[46]
Advances in Neural Information Processing Systems (NeurIPS) , year=
Angular Steering: Behavior Control via Rotation in Activation Space , author=. Advances in Neural Information Processing Systems (NeurIPS) , year=
-
[47]
Findings of the Association for Computational Linguistics: EMNLP 2025 , pages=
Beyond Linear Steering: Unified Multi-Attribute Control for Language Models , author=. Findings of the Association for Computational Linguistics: EMNLP 2025 , pages=. 2025 , publisher=
2025
-
[48]
ICLR Workshop on Building Trust in
Interpretable Steering of Large Language Models with Feature Guided Activation Additions , author=. ICLR Workshop on Building Trust in. 2025 , note=
2025
-
[49]
IEEE Access , volume=
The Effectiveness of Style Vectors for Steering Large Language Models: A Human Evaluation , author=. IEEE Access , volume=. 2025 , note=
2025
-
[50]
Proceedings of the 7th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP , pages=
Multi-property Steering of Large Language Models with Dynamic Activation Composition , author=. Proceedings of the 7th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP , pages=. 2024 , note=
2024
-
[51]
Adaptive Activation Steering: A Tuning-Free
Wang, Tianlong and Jiao, Xianfeng and Zhu, Yinghao and Chen, Zhongzhi and He, Yifan and Chu, Xu and Gao, Junyi and Wang, Yasha and Ma, Liantao , booktitle=. Adaptive Activation Steering: A Tuning-Free. 2025 , note=
2025
-
[52]
Contributions to the Theory of Games II , editor=
A Value for n-Person Games , author=. Contributions to the Theory of Games II , editor=. 1953 , publisher=
1953
-
[53]
Zhang, Yanzhao and Li, Mingxin and Long, Dingkun and Zhang, Xin and Lin, Huan and Yang, Baosong and Xie, Pengjun and Yang, An and Liu, Dayiheng and Lin, Junyang and Huang, Fei and Zhou, Jingren , journal=
-
[54]
Cumulated Gain-Based Evaluation of
J. Cumulated Gain-Based Evaluation of. ACM Transactions on Information Systems (TOIS) , volume=
Reviewed August 14, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.