mlsanityData to 14 Oct 2026

SAE features and their applications

231 papers4 this weekMedium confidence in this grouping: mostly one topic, with blurry edgesMedium confidence in this grouping: mostly one topic, with blurry edges · About

Keywords: features, sparse, saes, sae, feature

Key facts. SAE features and their applications: 231 papers on arXiv since 2022 (data through 14 Oct 2026). 171 of its 231 mapped papers (74%) cite “Sparse Autoencoders Find Highly Interpretable Features in Language Models” (2023, arXiv 2309.08600). 13 new papers 24 Sep–14 Oct.

What’s new

24 Sep – 14 Oct 2026

13 new papers, 24 Sep – 14 Oct; 4 this week.

  1. This week · 4
  2. Diminishing Returns of Data in Sparse Autoencoders: Functional Behavior does not Guarantee Feature Reproducibility also fits: Sparse autoencoder training and evaluation
  3. Sparse Autoencoders for Mechanistic Interpretability in NLP: A Critical Survey of Recoverability, Interpretability, and Utility
  4. Sparse Autoencoders Are a Shared Coordinate System for Symbolic Matching
  5. Evaluating the Causal Faithfulness of SHAP as a Candidate Attribution Method for SAE Features also fits: Probing deception, bias and beliefs
  6. Earlier · 9
  7. Disentangling Linguistic and Paralinguistic Information with Routed Sparse Autoencoders
  8. A Testable Theory of Atomic Features
  9. Active Budget Can Kill Sensitivity: Diagnosing and Repairing TopK Sparse Autoencoder Reliability also fits: Sparse autoencoder training and evaluation
  10. When Trees Are Not Enough: Learning Mixed-Topology Feature Graphs with Adaptive Graph Sparse Autoencoders
  11. Beyond Token Scale: Chunk-Level Sparse Autoencoders for Reliable Semantic Feature Discovery
  12. From Input to Output: A Flexible Agent for Dual-End Interpretation of Sparse Autoencoder Features also fits: Sparse autoencoder training and evaluation
All 13
  1. Routing Without Embeddings: Fast And Interpretable Routing With Regular Expressions
  2. Gradients for Interventions and Activations for Detection: Targeted Feature Learning in Language Models
  3. Parts-of-Speech as Emergent Categories in SAE Latent Space

Foundations

Most cited by papers here

  1. Sparse Autoencoders Find Highly Interpretable Features in Language Models2023cited by 171 here
  2. Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoders2024cited by 87 here
  3. Scaling and evaluating sparse autoencoders2024cited by 82 here
  4. Improving Dictionary Learning with Gated Sparse Autoencoders2024cited by 51 here
  5. A is for Absorption: Studying Feature Splitting and Absorption in Sparse Autoencoders2024cited by 46 here

Lineage

The papers this direction builds on, oldest first.

Each paper on the line cites the one before.

  1. 2022Toy Models of Superpositioncited by 63 here
  2. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 171 here
  3. 2024Improving Dictionary Learning with Gated Sparse Autoencoderscited by 51 here
  4. 2024Identifying Functionally Important Features with End-to-End Sparse Dictionary Learningcited by 21 here
  5. 2024Scaling and evaluating sparse autoencoderscited by 82 here
  6. 2024Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoderscited by 87 here
  7. 2024Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2cited by 103 here
  8. 2024A is for Absorption: Studying Feature Splitting and Absorption in Sparse Autoencoderscited by 46 here
  9. 2025SAEBench: A Comprehensive Benchmark for Sparse Autoencoders in Language Model Interpretabilitycited by 43 here
  10. 2025Learning Multi-Level Features with Matryoshka Sparse Autoencoderscited by 30 here
14 more it builds on, not in this chain
  1. 2022Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 smallcited by 25 here
  2. 2023Steering Language Models With Activation Engineeringcited by 22 here
  3. 2023Steering Llama 2 via Contrastive Activation Additioncited by 33 here
  4. 2024Mechanistic Interpretability for AI Safety -- A Reviewcited by 27 here
  5. 2024Towards Principled Evaluations of Sparse Autoencoders for Interpretability and Controlcited by 24 here
  6. 2024Transcoders Find Interpretable LLM Feature Circuitscited by 21 here
  7. 2024Automatically Interpreting Millions of Features in Large Language Modelscited by 33 here
  8. 2024Applying sparse autoencoders to unlearn knowledge in language modelscited by 21 here
  9. 2024Llama Scope: Extracting Millions of Features from Llama-3.1-8B with Sparse Autoencoderscited by 45 here
  10. 2024Improving Steering Vectors by Targeting Sparse Autoencoder Featurescited by 21 here
  11. 2024Do I Know This Entity? Knowledge Awareness and Hallucinations in Language Modelscited by 19 here
  12. 2025AxBench: Steering LLMs? Even Simple Baselines Outperform Sparse Autoencoderscited by 27 here
  13. 2025Sparse Autoencoders Do Not Find Canonical Units of Analysiscited by 22 here
  14. 2025SAEs Are Good for Steering -- If You Select the Right Featurescited by 21 here

The 24 most cited of 274.

Browse

30 of 231, ranked 6–35

  1. 6Llama Scope: Extracting Millions of Features from Llama-3.1-8B with Sparse Autoencoders202445
  2. 7SAEBench: A Comprehensive Benchmark for Sparse Autoencoders in Language Model Interpretability202543
  3. 8Automatically Interpreting Millions of Features in Large Language Models202433
  4. 9AxBench: Steering LLMs? Even Simple Baselines Outperform Sparse Autoencoders202527
  5. 10SAEs Are Good for Steering -- If You Select the Right Features202521
  6. 11Applying sparse autoencoders to unlearn knowledge in language models202421
  7. 12Improving Steering Vectors by Targeting Sparse Autoencoder Features202421
  8. 13Sparse Autoencoders Trained on the Same Data Learn Different Features202519
  9. 14Are Sparse Autoencoders Useful? A Case Study in Sparse Probing202519
  10. 15Measuring Progress in Dictionary Learning for Language Model Interpretability with Board Game Models202418
All 30
  1. 16BatchTopK Sparse Autoencoders202418
  2. 17A Survey on Sparse Autoencoders: Interpreting the Internal Mechanisms of Large Language Models202515
  3. 18Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet202614
  4. 19SAIF: A Sparse Autoencoder Framework for Interpreting and Steering Instruction Following of Language Models202511
  5. 20Quantifying Feature Space Universality Across Large Language Models via Sparse Autoencoders202410
  6. 21Steering Knowledge Selection Behaviours in LLMs via SAE-Based Representation Engineering202410
  7. 22Sparse Autoencoder Features for Classifications and Transferability20258
  8. 23SAE-SSV: Supervised Steering in Sparse Representation Spaces for Reliable Control of Language Models20258
  9. 24Interpreting and Steering LLMs with Mutual Information-based Explanations on Sparse Autoencoders20256
  10. 25Route Sparse Autoencoder to Interpret Large Language Models20256
  11. 26SAEs Can Improve Unlearning: Dynamic Sparse Autoencoder Guardrails for Precision Unlearning in LLMs20256
  12. 27Dense SAE Latents Are Features, Not Bugs20256
  13. 28Rethinking Evaluation of Sparse Autoencoders through the Representation of Polysemous Words20255
  14. 29Unveiling Language-Specific Features in Large Language Models via Sparse Autoencoders20255
  15. 30Can sparse autoencoders be used to decompose and interpret steering vectors?20245
  16. 31FADE: Why Bad Descriptions Happen to Good Features20254
  17. 32Causal Language Control in Multilingual Transformers via Sparse Feature Steering20254
  18. 33Residual Stream Analysis with Multi-Layer SAEs20244
  19. 34Evolution of SAE Features Across Layers in LLMs20244
  20. 35Group-SAE: Efficient Training of Sparse Autoencoders for Large Language Models via Layer Groups20244