mlsanityData to 7 Oct 2026

Interpretable Steering of Large Language Models with Feature Guided Activation Additions

arXiv 2501.09929 ↗First posted 17 Jan 2025cs.LG cs.AI cs.CLcited by 32

Not on the map. Why?

Abstract (arXiv)

Effective and reliable control over large language model (LLM) behavior is a significant challenge. While activation steering methods, which add steering vectors to a model's hidden states, are a promising approach, existing techniques often lack precision and interpretability in how they influence model outputs. We introduce Feature Guided Activation Additions (FGAA), a novel activation steering method that leverages insights from Contrastive Activation Addition (CAA) and Sparse Autoencoder-Targeted Steering (SAE-TS). By operating in the latent space of a Sparse Autoencoder (SAE) and employing optimization techniques to select desired SAE features, FGAA constructs precise steering vectors that provide better steering effects while maintaining coherence of steered model outputs. In this regard, evaluations on Gemma-2-2B and Gemma-2-9B models across various steering tasks demonstrate that FGAA outperforms existing steering methods of CAA, SAE decoder steering, and SAE-TS. Our results also highlight important trade-offs between steering scale and general model capabilities that are consistent across all tested steering methods.

Key facts. Cited by 32 papers in the map (data through 7 Oct 2026). Builds on 16 of its 24 references that are in the map. First posted on arXiv 17 Jan 2025.

Builds on

16 of 24 references.

References · 16

  1. 2022Training language models to follow instructions with human feedbackcited by 15,712
  2. 2020Measuring Massive Multitask Language Understandingcited by 7,768
  3. 2024MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmarkcited by 1,758
  4. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 1,359
  5. 2023Steering Llama 2 via Contrastive Activation Additioncited by 1,009
  6. 2022Toy Models of Superpositioncited by 928
  7. 2023Steering Language Models With Activation Engineeringcited by 824
  8. 2023The Linear Representation Hypothesis and the Geometry of Large Language Modelscited by 679
  9. 2024The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructionscited by 441
  10. 2024Mechanistic Interpretability for AI Safety -- A Reviewcited by 415
  11. 2024Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2cited by 381
  12. 2024A is for Absorption: Studying Feature Splitting and Absorption in Sparse Autoencoderscited by 148
4 more
  1. 2024Automatically Interpreting Millions of Features in Large Language Modelscited by 106
  2. 2023NEFTune: Noisy Embeddings Improve Instruction Finetuningcited by 99
  3. 2024Improving Steering Vectors by Targeting Sparse Autoencoder Featurescited by 86
  4. 2022Polysemanticity and Capacity in Neural Networkscited by 74

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 14; 2026: 18.

32 papers.

Sort:

By area:

  1. 2025I Have Covered All the Bases Here: Interpreting Reasoning Features in Large Language Models via Sparse Autoencoderscited by 43
  2. 2026Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Modelscited by 22
  3. 2025The Rogue Scalpel: Activation Steering Compromises LLM Safetycited by 18
  4. 2026Not All Features Are Created Equal: A Mechanistic Study of Vision-Language-Action Modelscited by 14
  5. 2025Causal Language Control in Multilingual Transformers via Sparse Feature Steeringcited by 12
  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. 2025Enhancing LLM Steering through Sparse Autoencoder-Based Vector Refinementcited by 9
  9. 2025The Geometry of Harmfulness in LLMs through Subconcept Probingcited by 7
  10. 2025What Can We Actually Steer? A Multi-Behavior Study of Activation Controlcited by 6
  11. 2026Steering at the Source: Style Modulation Heads for Robust Persona Controlcited by 4
  12. 2026Multilingual Steering by Design: Multilingual Sparse Autoencoders and Principled Layer Selectioncited by 4
20 more
  1. 2025Beyond Input Activations: Identifying Influential Latents by Gradient Sparse Autoencoderscited by 3
  2. 2025Bottlenecked Transformers: Periodic KV Cache Consolidation for Generalised Reasoningcited by 3
  3. 2026Abstract Activation Spaces for Content-Invariant Reasoning in Large Language Modelscited by 3
  4. 2025Steering Information Utility in Key-Value Memory for Language Model Post-Trainingcited by 2
  5. 2025Painless Activation Steering: An Automated, Lightweight Approach for Post-Training Large Language Modelscited by 2
  6. 2026Deployable Per-Instance Multi-Layer Activation Steering for Large Language Modelscited by 2
  7. 2025Fusion Steering: Prompt-Specific Activation Controlcited by 1
  8. 2026Steerability of Instrumental-Convergence Tendencies in LLMscited by 1
  9. 2026Prometheus Mind: Retrofitting Memory to Frozen Language Modelscited by 1
  10. 2026Mechanistic Interpretability for Neural Networks: Circuits, Sparse Features and Symbolic Reasoningcited by 1
  11. 2025Beyond topography: Topographic regularization improves robustness and reshapes representations in convolutional neural networkscited by 0
  12. 2026Mechanistic Indicators of Steering Effectiveness in Large Language Modelscited by 0
  13. 2026Do Language Models Encode Semantic Relations? Probing and Sparse Feature Analysiscited by 0
  14. 2026Continuous Interpretive Steering for Scalar Diversitycited by 0
  15. 2026Why Are Some Emotions Harder for LLMs? Uncovering the Causal Mechanisms of Emotion Inference via Sparse Autoencoderscited by 0
  16. 2026REALISTA: Realistic Latent Adversarial Attacks that Elicit LLM Hallucinationscited by 0
  17. 2026Transcoders Trace Visual Grounding and Hallucinations in Vision-Language Modelscited by 0
  18. 2026Steered Generation via Gradient-Based Optimization on Sparse Query Featurescited by 0
  19. 2026Measuring Semantic Abstractness of SAE Features via Nonlocalitycited by 0
  20. 2026MetaSteer: Context-Conditioned, nonlinear Steering via Attention-Projection Adaptationcited by 0