mlsanityData to 7 Oct 2026

Improving Steering Vectors by Targeting Sparse Autoencoder Features

arXiv 2411.02193 ↗First posted 4 Nov 2024cs.LG cs.AI cs.CLcited by 86

Not on the map. Why?

Abstract (arXiv)

To control the behavior of language models, steering methods attempt to ensure that outputs of the model satisfy specific pre-defined properties. Adding steering vectors to the model is a promising method of model control that is easier than finetuning, and may be more robust than prompting. However, it can be difficult to anticipate the effects of steering vectors produced by methods such as CAA [Panickssery et al., 2024] or the direct use of SAE latents [Templeton et al., 2024]. In our work, we address this issue by using SAEs to measure the effects of steering vectors, giving us a method that can be used to understand the causal effect of any steering vector intervention. We use this method for measuring causal effects to develop an improved steering method, SAE-Targeted Steering (SAE-TS), which finds steering vectors to target specific SAE features while minimizing unintended side effects. We show that overall, SAE-TS balances steering effects with coherence better than CAA and SAE feature steering, when evaluated on a range of tasks.

Key facts. Cited by 86 papers in the map (data through 7 Oct 2026). Builds on 14 of its 40 references that are in the map. First posted on arXiv 4 Nov 2024.

Builds on

14 of 40 references.

References · 14

  1. 2022Training language models to follow instructions with human feedbackcited by 15,712
  2. 2023Jailbroken: How Does LLM Safety Training Fail?cited by 1,565
  3. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 1,359
  4. 2023Representation Engineering: A Top-Down Approach to AI Transparencycited by 1,239
  5. 2023Steering Llama 2 via Contrastive Activation Additioncited by 1,009
  6. 2022Toy Models of Superpositioncited by 928
  7. 2024The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructionscited by 441
  8. 2024Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Modelscited by 412
  9. 2024Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2cited by 381
  10. 2024Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoderscited by 283
  11. 2024Programming Refusal with Conditional Activation Steeringcited by 191
  12. 2016Linear Algebraic Structure of Word Senses, with Applications to Polysemycited by 165
2 more
  1. 2024Personalized Steering of Large Language Models: Versatile Steering Vectors Through Bi-directional Preference Optimizationcited by 112
  2. 2024Steering Knowledge Selection Behaviours in LLMs via SAE-Based Representation Engineeringcited by 60

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 2; 2025: 43; 2026: 41.

86 papers.

Sort:

By area:

  1. 2025AxBench: Steering LLMs? Even Simple Baselines Outperform Sparse Autoencoderscited by 199
  2. 2025Are Sparse Autoencoders Useful? A Case Study in Sparse Probingcited by 109
  3. 2024Steering Language Model Refusal with Sparse Autoencoderscited by 89
  4. 2025Persona Features Control Emergent Misalignmentcited by 76
  5. 2025A Survey on Sparse Autoencoders: Interpreting the Internal Mechanisms of Large Language Modelscited by 73
  6. 2025Enhancing Automated Interpretability with Output-Centric Feature Descriptionscited by 39
  7. 2025Interpretable Steering of Large Language Models with Feature Guided Activation Additionscited by 32
  8. 2025SAE-SSV: Supervised Steering in Sparse Representation Spaces for Reliable Control of Language Modelscited by 28
  9. 2025Beyond Prompt Engineering: Robust Behavior Control in LLMs via Steering Target Atomscited by 23
  10. 2025Representation Engineering for Large-Language Models: Survey and Research Challengescited by 18
  11. 2024Tracking the Feature Dynamics in LLM Training: A Mechanistic Studycited by 15
  12. 2025Does Time Have Its Place? Temporal Heads: Where Language Models Recall Time-specific Informationcited by 14
38 more
  1. 2025Position: Mechanistic Interpretability Should Prioritize Feature Consistency in SAEscited by 14
  2. 2025Analyze Feature Flow to Enhance Interpretation and Steering in Language Modelscited by 12
  3. 2025SAUCE: Selective Concept Unlearning in Vision-Language Models with Sparse Autoencoderscited by 12
  4. 2025Causal Language Control in Multilingual Transformers via Sparse Feature Steeringcited by 12
  5. 2025Evaluating Adversarial Robustness of Concept Representations in Sparse Autoencoderscited by 11
  6. 2025Interpretation Meets Safety: A Survey on Interpretation Methods and Tools for Improving LLM Safetycited by 10
  7. 2025CorrSteer: Generation-Time LLM Steering via Correlated Sparse Autoencoder Featurescited by 10
  8. 2025Does higher interpretability imply better utility? A Pairwise Analysis on Sparse Autoencoderscited by 9
  9. 2026Analysing the Safety Pitfalls of Steering Vectorscited by 9
  10. 2025Mitigating Memorization in LLMs using Activation Steeringcited by 7
  11. 2025EasyEdit2: An Easy-to-use Steering Framework for Editing Large Language Modelscited by 7
  12. 2025Interpreting Large Text-to-Image Diffusion Models with Dictionary Learningcited by 6
  13. 2025Line of Sight: On Linear Representations in VLLMscited by 6
  14. 2025Neurons Speak in Ranges: Breaking Free from Discrete Neuronal Attributioncited by 5
  15. 2026Steering Safely or Off a Cliff? Rethinking Specificity and Robustness in Inference-Time Interventionscited by 5
  16. 2026The Geometry of Representational Failures in Vision Language Modelscited by 5
  17. 2026Stop Probing, Start Coding: Why Linear Probes and Sparse Autoencoders Fail at Compositional Generalisationcited by 5
  18. 2025Steering Pretrained Drafters during Speculative Decodingcited by 4
  19. 2025AlignSAE: Concept-Aligned Sparse Autoencoderscited by 4
  20. 2025Depth-Wise Activation Steering for Honest Language Modelscited by 4
  21. 2026SegCompass: Exploring Interpretable Alignment with Sparse Autoencoders for Enhanced Reasoning Segmentationcited by 4
  22. 2026Multilingual Steering by Design: Multilingual Sparse Autoencoders and Principled Layer Selectioncited by 4
  23. 2026Perplexity Can Miss SAE Feature Damage Under Quantizationcited by 4
  24. 2025Beyond Input Activations: Identifying Influential Latents by Gradient Sparse Autoencoderscited by 3
  25. 2025Bottlenecked Transformers: Periodic KV Cache Consolidation for Generalised Reasoningcited by 3
  26. 2025But what is your honest answer? Aiding LLM-judges with honest alternatives using steering vectorscited by 3
  27. 2025Self-Improving Model Steeringcited by 3
  28. 2025The Anatomy of Alignment: Decomposing Preference Optimization by Steering Sparse Featurescited by 3
  29. 2025A Comparative Analysis of Sparse Autoencoder and Activation Difference in Language Model Steeringcited by 3
  30. 2025Transmuting prompts into weightscited by 3
  31. 2026Spectral Superposition: A Theory of Feature Geometrycited by 3
  32. 2026The Cylindrical Representation Hypothesis for Language Model Steeringcited by 3
  33. 2025Patterns and Mechanisms of Contrastive Activation Engineeringcited by 2
  34. 2025Steering Large Language Models for Machine Translation Personalizationcited by 2
  35. 2025Visual Exploration of Feature Relationships in Sparse Autoencoders with Curated Conceptscited by 2
  36. 2026YaPO: Learnable Sparse Activation Steering Vectors for Domain Adaptationcited by 2
  37. 2026SteerRM: Debiasing Reward Models via Sparse Autoencoderscited by 2
  38. 2026Tracing the Thought of a Grandmaster-level Chess-Playing Transformercited by 2

50 most cited of 86 All 86 (JSON)