mlsanityData to 14 Oct 2026

Can sparse autoencoders be used to decompose and interpret steering vectors?

arXiv 2411.08790 ↗First posted 13 Nov 2024cs.LG cs.AI cs.CLcited by 19

On the map: Mechanistic interpretability → SAE features and their applications

Abstract (arXiv)

Steering vectors are a promising approach to control the behaviour of large language models. However, their underlying mechanisms remain poorly understood. While sparse autoencoders (SAEs) may offer a potential method to interpret steering vectors, recent findings show that SAE-reconstructed vectors often lack the steering properties of the original vectors. This paper investigates why directly applying SAEs to steering vectors yields misleading decompositions, identifying two reasons: (1) steering vectors fall outside the input distribution for which SAEs are designed, and (2) steering vectors can have meaningful negative projections in feature directions, which SAEs are not designed to accommodate. These limitations hinder the direct use of SAEs for interpreting steering vectors.

Key facts. Cited by 19 papers in the map (data through 14 Oct 2026). Placed in SAE features and their applications, one of 13 directions in Mechanistic interpretability. Builds on 17 of its 23 references that are in the map. First posted on arXiv 13 Nov 2024.

Builds on

17 of 23 references.

Closest ancestors · 3

Same direction, newest first

  1. 2024Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoders19 Jul
  2. 2024Scaling and evaluating sparse autoencoders6 Jun
  3. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Models15 Sep

Other references · 14

  1. 2020The Pile: An 800GB Dataset of Diverse Text for Language Modelingcited by 2,411
  2. 2023Representation Engineering: A Top-Down Approach to AI Transparencycited by 1,336
  3. 2023Steering Llama 2 via Contrastive Activation Additioncited by 1,085
  4. 2024Refusal in Language Models Is Mediated by a Single Directioncited by 1,032
  5. 2023Steering Language Models With Activation Engineeringcited by 922
  6. 2022Discovering Language Model Behaviors with Model-Written Evaluationscited by 836
  7. 2024Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Modelscited by 435
  8. 2024Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2cited by 394
  9. 2023In-context Vectors: Making In Context Learning More Effective and Controllable Through Latent Space Steeringcited by 245
  10. 2024Foundational Challenges in Assuring Alignment and Safety of Large Language Modelscited by 199
  11. 2024Analyzing the Generalization and Reliability of Steering Vectorscited by 143
  12. 2024Extending Activation Steering to Broad Skills and Multiple Behaviourscited by 40
2 more
  1. 2024Model Editing as a Robust and Denoised variant of DPO: A Case Study on Toxicitycited by 35
  2. 2024Sparse Autoencoders Enable Scalable and Reliable Circuit Identification in Language Modelscited by 15

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 12; 2026: 7.

19 papers.

Sort:

By area:

  1. 2025AxBench: Steering LLMs? Even Simple Baselines Outperform Sparse Autoencoderscited by 218
  2. 2025Archetypal SAE: Adaptive and Stable Dictionary Learning for Concept Extraction in Large Vision Modelscited by 73
  3. 2025To Steer or Not to Steer? Mechanistic Error Reduction with Abstention for Language Modelscited by 29
  4. 2025Representation Engineering for Large-Language Models: Survey and Research Challengescited by 21
  5. 2025Unveiling Language-Specific Features in Large Language Models via Sparse Autoencoderscited by 13
  6. 2026Steering Vector Fields for Context-Aware Inference-Time Control in Large Language Modelscited by 13
  7. 2025Does higher interpretability imply better utility? A Pairwise Analysis on Sparse Autoencoderscited by 10
  8. 2026From Weights to Activations: Is Steering the Next Frontier of Adaptation?cited by 8
  9. 2025EasyEdit2: An Easy-to-use Steering Framework for Editing Large Language Modelscited by 7
  10. 2026DLM-Scope: Mechanistic Interpretability of Diffusion Language Models via Sparse Autoencoderscited by 7
  11. 2025Evaluate Bias without Manual Test Sets: A Concept Representation Perspective for LLMscited by 6
  12. 2025A Comparative Analysis of Sparse Autoencoder and Activation Difference in Language Model Steeringcited by 4
7 more
  1. 2025SASFT: Sparse Autoencoder-guided Supervised Finetuning to Mitigate Unexpected Code-Switching in LLMscited by 2
  2. 2025Rethinking Sparse Autoencoders: Select-and-Project for Fairness and Control from Encoder Features Alonecited by 1
  3. 2025Polarity-Aware Probing for Quantifying Latent Alignment in Language Modelscited by 1
  4. 2026Auditing Asset-Specific Preferences in Financial Large Language Models: Evidence from Bitcoin Representations and Portfolio Allocationcited by 1
  5. 2026Look But Don't Touch with Sparse Autoencoders for Unlearning in Diffusion Modelscited by 1
  6. 2026Beyond a Bag of Features: Set-Level Instability in Sparse Autoencoderscited by 0
  7. 2026Steering Fields: Adaptive Vector Fields for Safe Image Generation and Beyondcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Steering LLMs? Actually, Sparse Autoencoders can outperform simple baselines
  2. 2024Improving Steering Vectors by Targeting Sparse Autoencoder Features
  3. 2025Enhancing LLM Steering through Sparse Autoencoder-Based Vector Refinement
  4. 2025SAEs Are Good for Steering -- If You Select the Right Features
  5. 2026Pre-Intervention Prediction of Sparse Autoencoder Steering Side Effects
  6. 2025SAE-SSV: Supervised Steering in Sparse Representation Spaces for Reliable Control of Language Models
  7. 2025Beyond Interpretability: When, Why, and How Sparse Autoencoders Enable Label-Free Visual Steering
  8. 2026Are Sparse Autoencoder Benchmarks Reliable?