mlsanityData to 14 Oct 2026

Circuits and mechanistic analysis

272 papers1 this weekHigh confidence in this grouping: the papers form one clear topicHigh confidence in this grouping: the papers form one clear topic · About

Keywords: circuit, mechanistic, interpretability, circuits, mechanistic interpretability

Key facts. Circuits and mechanistic analysis: 272 papers on arXiv since 2022 (data through 14 Oct 2026). 148 of its 272 mapped papers (54%) cite “Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 small” (2022, arXiv 2211.00593). 10 new papers 24 Sep–14 Oct.

What’s new

24 Sep – 14 Oct 2026

10 new papers, 24 Sep – 14 Oct; 1 this week.

  1. This week · 1
  2. Interpreting Language Model Parameters also fits: Knowledge and representations inside LLMs
  3. Earlier · 9
  4. The Ball and the Box: Two Geometries of Computation in Superposition also fits: Sparse autoencoder training and evaluation
  5. CircuitTree: Composing Circuits into A Syllable Transcriber
  6. Slaying the Hydra: Interaction-Aware Circuit Discovery in Language Models
  7. Certified Mechanistic Edits: Behavioral Guarantees for Skill Removal and Preservation
  8. Are We Recovering Mechanisms? Objective-Level Recovery Gaps in Mechanistic Interpretability
  9. When Models Don't Manipulate Manifolds: The Geometry of a Comparison Task
  10. Rethinking Circuit Evaluation: Do Circuits Explain Model Errors?
  11. Representational Simplicity and Circuit Size Dissociate in a Threshold-Dependent Way: A Controlled Test via Adversarial Training also fits: Sparse autoencoder training and evaluation
  12. Mechanistic Interpretability Reveals Shared Causal Subspaces in Brain-to-Speech Decoders

Foundations

Most cited by papers here

  1. Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 small2022cited by 148 here
  2. Towards Automated Circuit Discovery for Mechanistic Interpretability2023cited by 128 here
  3. Progress measures for grokking via mechanistic interpretability2023cited by 68 here
  4. Attribution Patching Outperforms Automated Circuit Discovery2023cited by 63 here
  5. Mechanistic Interpretability for AI Safety -- A Review2024cited by 29 here

Lineage

The papers this direction builds on, oldest first.

Each paper on the line cites the one before.

  1. 2022Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 smallcited by 148 here
  2. 2023Progress measures for grokking via mechanistic interpretabilitycited by 68 here
  3. 2023A Toy Model of Universality: Reverse Engineering How Networks Learn Group Operationscited by 24 here
  4. 2023Towards Automated Circuit Discovery for Mechanistic Interpretabilitycited by 128 here
  5. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 75 here
  6. 2023Towards Best Practices of Activation Patching in Language Models: Metrics and Methodscited by 38 here
  7. 2024Transcoders Find Interpretable LLM Feature Circuitscited by 28 here
  8. 2024Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2cited by 14 here
  9. 2025MIB: A Mechanistic Interpretability Benchmarkcited by 20 here
15 more it builds on, not in this chain
  1. 2022Toy Models of Superpositioncited by 66 here
  2. 2023Causal Abstraction: A Theoretical Foundation for Mechanistic Interpretabilitycited by 26 here
  3. 2023Eliciting Latent Predictions from Transformers with the Tuned Lenscited by 13 here
  4. 2023Steering Language Models With Activation Engineeringcited by 17 here
  5. 2023Representation Engineering: A Top-Down Approach to AI Transparencycited by 26 here
  6. 2023Circuit Component Reuse Across Tasks in Transformer Language Modelscited by 19 here
  7. 2023Attribution Patching Outperforms Automated Circuit Discoverycited by 63 here
  8. 2023Is This the Subspace You Are Looking for? An Interpretability Illusion for Subspace Activation Patchingcited by 14 here
  9. 2023Steering Llama 2 via Contrastive Activation Additioncited by 12 here
  10. 2024AtP*: An efficient and scalable method for localizing LLM behaviour to componentscited by 28 here
  11. 2024Mechanistic Interpretability for AI Safety -- A Reviewcited by 29 here
  12. 2024Scaling and evaluating sparse autoencoderscited by 21 here
  13. 2024Efficient Automated Circuit Discovery in Transformers using Contextual Decompositioncited by 13 here
  14. 2024A Practical Review of Mechanistic Interpretability for Transformer-Based Language Modelscited by 16 here
  15. 2025Open Problems in Mechanistic Interpretabilitycited by 22 here

The 24 most cited of 243.

Browse

30 of 272, ranked 6–35

  1. 6Transcoders Find Interpretable LLM Feature Circuits202428
  2. 7Causal Abstraction: A Theoretical Foundation for Mechanistic Interpretability202326
  3. 8A Toy Model of Universality: Reverse Engineering How Networks Learn Group Operations202324
  4. 9Open Problems in Mechanistic Interpretability202522
  5. 10MIB: A Mechanistic Interpretability Benchmark202520
  6. 11Circuit Component Reuse Across Tasks in Transformer Language Models202319
  7. 12A Practical Review of Mechanistic Interpretability for Transformer-Based Language Models202416
  8. 13Is This the Subspace You Are Looking for? An Interpretability Illusion for Subspace Activation Patching202314
  9. 14Efficient Automated Circuit Discovery in Transformers using Contextual Decomposition202413
  10. 15Hypothesis Testing the Circuit Hypothesis in LLMs202412
All 30
  1. 16Transformer Circuit Faithfulness Metrics are not Robust202411
  2. 17Everything, Everywhere, All at Once: Is Mechanistic Interpretability Identifiable?202510
  3. 18Weight-sparse transformers have interpretable circuits202510
  4. 19Dictionary Learning Improves Patch-Free Circuit Discovery in Mechanistic Interpretability: A Case Study on Othello-GPT202410
  5. 20Automatically Identifying Local and Global Circuits with Linear Computation Graphs202410
  6. 21InterpBench: Semi-Synthetic Transformers for Evaluating Mechanistic Interpretability Techniques202410
  7. 22EAP-GP: Mitigating Saturation Effect in Gradient-based Automated Circuit Identification20259
  8. 23Universal Neurons in GPT2 Language Models20249
  9. 24RelP: Faithful and Efficient Circuit Discovery in Language Models via Relevance Patching20258
  10. 25Interpretability in Parameter Space: Minimizing Mechanistic Description Length with Attribution-based Parameter Decomposition20257
  11. 26Rethinking Circuit Completeness in Language Models: AND, OR, and ADDER Gates20257
  12. 27Compact Proofs of Model Performance via Mechanistic Interpretability20247
  13. 28Mechanistic Interpretability as Statistical Estimation: A Variance Analysis20256
  14. 29The Computational Complexity of Circuit Discovery for Inner Interpretability20246
  15. 30Towards Interpretable Sequence Continuation: Analyzing Shared Circuits in Large Language Models20236
  16. 31Position-aware Automatic Circuit Discovery20255
  17. 32The Non-Linear Representation Dilemma: Is Causal Abstraction Enough for Mechanistic Interpretability?20255
  18. 33PURE: Turning Polysemantic Neurons Into Pure Features by Identifying Relevant Circuits20245
  19. 34Disentangling Neuron Representations with Concept Vectors20235
  20. 35Seeing is Believing: Brain-Inspired Modular Training for Mechanistic Interpretability20235