mlsanityData to 7 Oct 2026

Towards Understanding Steering Strength

arXiv 2602.02712 ↗First posted 2 Feb 2026cs.LG cs.CLcited by 3

Not on the map. Why?

Abstract (arXiv)

A popular approach to post-training control of large language models (LLMs) is the steering of intermediate latent representations. Namely, identify a well-chosen direction depending on the task at hand and perturbs representations along this direction at inference time. While many propositions exist to pick this direction, considerably less is understood about how to choose the magnitude of the move, whereas its importance is clear: too little and the intended behavior does not emerge, too much and the model's performance degrades beyond repair. In this work, we propose the first theoretical analysis of steering strength. We characterize its effect on next token probability, presence of a concept, and cross-entropy, deriving precise qualitative laws governing these quantities. Our analysis reveals surprising behaviors, including non-monotonic effects of steering strength. We validate our theoretical predictions empirically on eleven language models, ranging from a small GPT architecture to modern models.

Key facts. Cited by 3 papers in the map (data through 7 Oct 2026). Builds on 37 of its 51 references that are in the map. First posted on arXiv 2 Feb 2026.

Builds on

37 of 51 references.

References · 37

  1. 2017Attention Is All You Needcited by 59,932
  2. 2023Llama 2: Open Foundation and Fine-Tuned Chat Modelscited by 12,865
  3. 2025Qwen3 Technical Reportcited by 8,195
  4. 2020Measuring Massive Multitask Language Understandingcited by 7,768
  5. 2016Layer Normalizationcited by 5,579
  6. 2021Finetuned Language Models Are Zero-Shot Learnerscited by 3,747
  7. 2022Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedbackcited by 3,634
  8. 2017Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layercited by 3,413
  9. 2019Fine-Tuning Language Models from Human Preferencescited by 2,124
  10. 2025Gemma 3 Technical Reportcited by 1,686
  11. 2019Root Mean Square Layer Normalizationcited by 1,425
  12. 2016Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddingscited by 1,304
25 more
  1. 2024The FineWeb Datasets: Decanting the Web for the Finest Text Data at Scalecited by 1,104
  2. 2023Steering Llama 2 via Contrastive Activation Additioncited by 1,009
  3. 2024Refusal in Language Models Is Mediated by a Single Directioncited by 960
  4. 2022Toy Models of Superpositioncited by 928
  5. 2023Steering Language Models With Activation Engineeringcited by 824
  6. 2022Discovering Latent Knowledge in Language Models Without Supervisioncited by 733
  7. 2023The Linear Representation Hypothesis and the Geometry of Large Language Modelscited by 679
  8. 2023The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasetscited by 673
  9. 2024Scaling and evaluating sparse autoencoderscited by 530
  10. 2023Emergent Linear Representations in World Models of Self-Supervised Sequence Modelscited by 313
  11. 2023Language Models Represent Space and Timecited by 306
  12. 2025Persona Vectors: Monitoring and Controlling Character Traits in Language Modelscited by 303
  13. 2025AxBench: Steering LLMs? Even Simple Baselines Outperform Sparse Autoencoderscited by 199
  14. 2021Exploring Deep Neural Networks via Layer-Peeled Model: Minority Collapse in Imbalanced Trainingcited by 166
  15. 2020Neural collapse with unconstrained featurescited by 140
  16. 2024Analyzing the Generalization and Reliability of Steering Vectorscited by 130
  17. 2025From Flat to Hierarchical: Extracting Sparse Representations with Matching Pursuitcited by 45
  18. 2024Steering Without Side Effects: Improving Post-Deployment Control of Language Modelscited by 44
  19. 2024A Language Model's Guide Through Latent Spacecited by 43
  20. 2025Taxonomy, Opportunities, and Challenges of Representation Engineering for Large Language Modelscited by 29
  21. 2025To Steer or Not to Steer? Mechanistic Error Reduction with Abstention for Language Modelscited by 26
  22. 2024Implicit Geometry of Next-token Prediction: From Language Sparsity Patterns to Model Representationscited by 24
  23. 2024Implicit Optimization Bias of Next-Token Prediction in Linear Modelscited by 14
  24. 2025Dynamically Scaled Activation Steeringcited by 6
  25. 2025Geometry of Semantics in Next-Token Prediction: How Optimization Implicitly Organizes Linguistic Representationscited by 2

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 3.

3 papers.

  1. 2026Steered LLM Activations are Non-Surjectivecited by 7
  2. 2026MoRFI: Monotonic Sparse Autoencoder Feature Identificationcited by 0
  3. 2026Activation Steering Transfer to Agents: One Gain Ratio Does Not Identify Potency and Efficacycited by 0