mlsanityData to 14 Oct 2026

Measuring Semantic Abstractness of SAE Features via Nonlocality

arXiv 2608.10537 ↗First posted 11 Aug 2026cs.AI cs.LGcited by 0

On the map: LLM reasoning → Reasoning benchmarks and evaluation

also in 2 more
Abstract (arXiv)

Sparse autoencoders (SAEs) have helped uncover mechanistic explanations for LLM behaviours such as reasoning, jailbreaking etc., via understanding the corresponding task-relevant and causally effective features. To evaluate such mechanistic explanations, downstream studies must distinguish surface lexical features from genuinely high-level ones. However, neither an autointerp-based semantic description nor causal steering utility fully resolves the abstraction level of a feature. To this end, we introduce Feature Nonlocality (FNL), defined as the entropy of the normalized per-position influence on an SAE feature's activation. We report that FNL correlates with existing LLM-based proxy metrics of feature semantic abstractness, and successfully distinguishes context-dependent reasoning features from token-driven ones, correctly assigning the higher FNL to the contextual feature in 73--84% of randomly drawn pairs that consist of one contextual and one token-level feature. We demonstrate two downstream applications. We audit SAE-based features used for jailbreak mitigation and find surprisingly that most effective features are positional features with low FNL rather than genuinely recognizing harmful intents. We report that steering high-FNL features in DeepSeek-R1-Distill-Llama-8B improves MATH-500 accuracy by 4.6 points over the unsteered model and outperforms steering low-FNL features, though the gains are model-specific. We conclude that FNL provides an LLM-independent, label-free, correlational witness of the abstraction level of an SAE feature, with applications in evaluating mechanistic explanations as well as selecting features for downstream interventions.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning. Builds on 24 of its 42 references that are in the map. First posted on arXiv 11 Aug 2026.

Builds on

24 of 42 references.

Closest ancestors · 3

Same direction, newest first

  1. 2026Do Sparse Autoencoders Identify Reasoning Features in Language Models?9 Jan
  2. 2025OpenThoughts: Data Recipes for Reasoning Models4 Jun
  3. 2025I Have Covered All the Bases Here: Interpreting Reasoning Features in Large Language Models via Sparse Autoencoders24 Mar

Other references · 21

  1. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  2. 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,371
  3. 2023Let's Verify Step by Stepcited by 3,789
  4. 2016Pointer Sentinel Mixture Modelscited by 3,096
  5. 2020The Pile: An 800GB Dataset of Diverse Text for Language Modelingcited by 2,411
  6. 2023Efficient Streaming Language Models with Attention Sinkscited by 2,186
  7. 2024MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmarkcited by 1,848
  8. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 1,420
  9. 2023Inference-Time Intervention: Eliciting Truthful Answers from a Language Modelcited by 1,302
  10. 2023Steering Llama 2 via Contrastive Activation Additioncited by 1,085
  11. 2024Refusal in Language Models Is Mediated by a Single Directioncited by 1,032
  12. 2023Steering Language Models With Activation Engineeringcited by 922
9 more
  1. 2025Persona Vectors: Monitoring and Controlling Character Traits in Language Modelscited by 345
  2. 2026Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnetcited by 191
  3. 2023A mathematical perspective on Transformerscited by 164
  4. 2024Llama Scope: Extracting Millions of Features from Llama-3.1-8B with Sparse Autoencoderscited by 143
  5. 2023Rigorously Assessing Natural Language Explanations of Neuronscited by 45
  6. 2025Interpretable Steering of Large Language Models with Feature Guided Activation Additionscited by 35
  7. 2025SAE-SSV: Supervised Steering in Sparse Representation Spaces for Reliable Control of Language Modelscited by 29
  8. 2026Controllable LLM Reasoning via Sparse Autoencoder-Based Steeringcited by 15
  9. 2026Agentic Publication Protocol: An Attempt to Modernize Scientific Publicationcited by 2

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Descriptive Collision in Sparse Autoencoder Auto-Interpretability: When One Explanation Describes Many Features
  2. 2025Measuring Sparse Autoencoder Feature Sensitivity
  3. 2026Beyond Activation Patterns: A Weight-Based Out-of-Context Explanation of Sparse Autoencoder Features
  4. 2024Quantifying Feature Space Universality Across Large Language Models via Sparse Autoencoders
  5. 2025Measuring and Guiding Monosemanticity
  6. 2026Beyond a Bag of Features: Set-Level Instability in Sparse Autoencoders
  7. 2025A Survey on Sparse Autoencoders: Interpreting the Internal Mechanisms of Large Language Models
  8. 2026Sparse Autoencoders for Mechanistic Interpretability in NLP: A Critical Survey of Recoverability, Interpretability, and Utility