mlsanityData to 7 Oct 2026

Towards Automated Circuit Discovery for Mechanistic Interpretability

arXiv 2304.14997 ↗First posted 28 Apr 2023cs.LGcited by 700

Not on the map. Why?

Cited by 34 in LLM reasoning, but placed by its own abstract, which didn’t fit.

Abstract (arXiv)

Through considerable effort and intuition, several recent works have reverse-engineered nontrivial behaviors of transformer models. This paper systematizes the mechanistic interpretability process they followed. First, researchers choose a metric and dataset that elicit the desired model behavior. Then, they apply activation patching to find which abstract neural network units are involved in the behavior. By varying the dataset, metric, and units under investigation, researchers can understand the functionality of each component. We automate one of the process' steps: to identify the circuit that implements the specified behavior in the model's computational graph. We propose several algorithms and reproduce previous interpretability results to validate them. For example, the ACDC algorithm rediscovered 5/5 of the component types in a circuit in GPT-2 Small that computes the Greater-Than operation. ACDC selected 68 of the 32,000 edges in GPT-2 Small, all of which were manually found by previous work. Our code is available at https://github.com/ArthurConmy/Automatic-Circuit-Discovery.

Key facts. Cited by 700 papers in the map (data through 7 Oct 2026). Builds on 56 of its 100 references that are in the map. First posted on arXiv 28 Apr 2023.

Builds on

56 of 100 references.

References · 56

  1. 2017Attention Is All You Needcited by 59,932
  2. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,204
  3. 2023GPT-4 Technical Reportcited by 18,132
  4. 2020Scaling Laws for Neural Language Modelscited by 6,411
  5. 2016"Why Should I Trust You?": Explaining the Predictions of Any Classifiercited by 5,040
  6. 2017Axiomatic Attribution for Deep Networkscited by 3,350
  7. 2016Categorical Reparameterization with Gumbel-Softmaxcited by 2,708
  8. 2022Locating and Editing Factual Associations in GPTcited by 2,609
  9. 2022Emergent Abilities of Large Language Modelscited by 2,483
  10. 2013Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Mapscited by 2,348
  11. 2015Learning both Weights and Connections for Efficient Neural Networkscited by 1,998
  12. 2017Towards A Rigorous Science of Interpretable Machine Learningcited by 1,400
38 more
  1. 2020Transformer Feed-Forward Layers Are Key-Value Memoriescited by 1,307
  2. 2018Sanity Checks for Saliency Mapscited by 1,149
  3. 2022Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 smallcited by 1,130
  4. 2023SparseGPT: Massive Language Models Can Be Accurately Pruned in One-Shotcited by 1,090
  5. 2022In-context Learning and Induction Headscited by 930
  6. 2022Toy Models of Superpositioncited by 928
  7. 2023Progress measures for grokking via mechanistic interpretabilitycited by 822
  8. 2019Are Sixteen Heads Really Better than One?cited by 803
  9. 2016The Mythos of Model Interpretabilitycited by 716
  10. 2017Interpretable Explanations of Black Boxes by Meaningful Perturbationcited by 611
  11. 2020Towards Faithfully Interpretable NLP Systems: How should we define and evaluate faithfulness?cited by 590
  12. 2017Learning Sparse Neural Networks through L0 Regularizationcited by 571
  13. 2022Hungry Hungry Hippos: Towards Language Modeling with State Space Modelscited by 495
  14. 2021Causal Abstractions of Neural Networkscited by 401
  15. 2020Movement Pruning: Adaptive Sparsity by Fine-Tuningcited by 398
  16. 2020A Survey on Neural Network Interpretabilitycited by 291
  17. 2019What's Hidden in a Randomly Weighted Neural Network?cited by 236
  18. 2019Structured Pruning of Large Language Modelscited by 201
  19. 2015Visualizing and Understanding Neural Models in NLPcited by 199
  20. 2020Causal Mediation Analysis for Interpreting Neural NLP: The Case of Gender Biascited by 183
  21. 2021Thinking Like Transformerscited by 182
  22. 2023Localizing Model Behavior with Path Patchingcited by 182
  23. 2020Compositional Explanations of Neuronscited by 172
  24. 2022A Fast Post-Training Pruning Framework for Transformerscited by 161
  25. 2018The Malicious Use of Artificial Intelligence: Forecasting, Prevention, and Mitigationcited by 159
  26. 2023A Toy Model of Universality: Reverse Engineering How Networks Learn Group Operationscited by 141
  27. 2022Natural Language Descriptions of Deep Visual Featurescited by 136
  28. 2022Toward Transparent AI: A Survey on Interpreting the Inner Structures of Deep Neural Networkscited by 126
  29. 2020CausaLM: Causal Model Explanation Through Counterfactual Language Modelscited by 117
  30. 2022The Optimal BERT Surgeon: Scalable and Accurate Second-Order Pruning for Large Language Modelscited by 115
  31. 2022Causal Machine Learning: A Survey and Open Problemscited by 107
  32. 2020Rewriting a Deep Generative Modelcited by 105
  33. 2023Tracr: Compiled Transformers as a Laboratory for Interpretabilitycited by 99
  34. 2021An Interpretability Illusion for BERTcited by 94
  35. 2019Analysis of Explainers of Black Box Deep Neural Networks for Computer Vision: A Surveycited by 93
  36. 2021Low-Complexity Probing via Finding Subnetworkscited by 63
  37. 2022X-Risk Analysis for AI Researchcited by 60
  38. 2019Analyzing and Interpreting Neural Networks for NLP: A Report on the First BlackboxNLP Workshopcited by 22

50 most cited of 56 All 56 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2023: 43; 2024: 152; 2025: 184; 2026: 321.

700 papers.

Sort:

By area:

  1. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 1,359
  2. 2024Mechanistic Interpretability for AI Safety -- A Reviewcited by 415
  3. 2024Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Modelscited by 412
  4. 2024Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2cited by 381
  5. 2023Towards Best Practices of Activation Patching in Language Models: Metrics and Methodscited by 328
  6. 2024Do Llamas Work in English? On the Latent Language of Multilingual Transformerscited by 275
  7. 2023Challenges and Applications of Large Language Modelscited by 251
  8. 2024Patchscopes: A Unifying Framework for Inspecting Hidden Representations of Language Modelscited by 225
  9. 2024Not All Language Model Features Are One-Dimensionally Linearcited by 201
  10. 2023Attribution Patching Outperforms Automated Circuit Discoverycited by 196
  11. 2024How to use and interpret activation patchingcited by 191
  12. 2024Transcoders Find Interpretable LLM Feature Circuitscited by 186
38 more
  1. 2024A Comprehensive Study of Knowledge Editing for Large Language Modelscited by 157
  2. 2023The Clock and the Pizza: Two Stories in Mechanistic Explanation of Neural Networkscited by 153
  3. 2023Does Circuit Analysis Interpretability Scale? Evidence from Multiple Choice Capabilities in Chinchillacited by 145
  4. 2023The Hydra Effect: Emergent Self-repair in Language Model Computationscited by 145
  5. 2024Have Faith in Faithfulness: Going Beyond Circuit Overlap When Finding Model Mechanismscited by 144
  6. 2024Fine-Tuning Enhances Existing Mechanisms: A Case Study on Entity Trackingcited by 140
  7. 2025Agentic Misalignment: How LLMs Could Be Insider Threatscited by 136
  8. 2024A Practical Review of Mechanistic Interpretability for Transformer-Based Language Modelscited by 113
  9. 2023Circuit Component Reuse Across Tasks in Transformer Language Modelscited by 105
  10. 2024Interpreting and Editing Vision-Language Representations to Mitigate Hallucinationscited by 103
  11. 2024Universal Neurons in GPT2 Language Modelscited by 100
  12. 2024Information Flow Routes: Automatically Interpreting Language Models at Scalecited by 92
  13. 2024A Primer on the Inner Workings of Transformer-based Language Modelscited by 83
  14. 2024Hopping Too Late: Exploring the Limitations of Large Language Models on Multi-Hop Queriescited by 78
  15. 2023Successor Heads: Recurring, Interpretable Attention Heads In The Wildcited by 77
  16. 2024RAVEL: Evaluating Interpretability Methods on Disentangling Language Model Representationscited by 77
  17. 2023Is This the Subspace You Are Looking for? An Interpretability Illusion for Subspace Activation Patchingcited by 75
  18. 2023Copy Suppression: Comprehensively Understanding an Attention Headcited by 73
  19. 2024On the Role of Attention Heads in Large Language Model Safetycited by 70
  20. 2024What's in the Image? A Deep-Dive into the Vision of Vision Language Modelscited by 66
  21. 2024Attention Heads of Large Language Models: A Surveycited by 65
  22. 2023Towards Vision-Language Mechanistic Interpretability: A Causal Tracing Tool for BLIPcited by 62
  23. 2024A Multimodal Automated Interpretability Agentcited by 62
  24. 2023Explaining black box text modules in natural language with language modelscited by 61
  25. 2024Knowledge Circuits in Pretrained Transformerscited by 60
  26. 2024LLM Circuit Analyses Are Consistent Across Training and Scalecited by 60
  27. 2024Finding Transformer Circuits with Edge Pruningcited by 58
  28. 2024Confidence Regulation Neurons in Language Modelscited by 57
  29. 2024Emergence of a High-Dimensional Abstraction Phase in Language Transformerscited by 56
  30. 2025Interpreting CLIP with Hierarchical Sparse Autoencoderscited by 54
  31. 2024pyvene: A Library for Understanding and Improving PyTorch Models via Interventionscited by 53
  32. 2024The Geometry of Concepts: Sparse Autoencoder Feature Structurecited by 52
  33. 2024From Yes-Men to Truth-Tellers: Addressing Sycophancy in Large Language Models with Pinpoint Tuningcited by 51
  34. 2024Knowledge Mechanisms in Large Language Models: A Survey and Perspectivecited by 50
  35. 2024A Mechanistic Analysis of a Transformer Trained on a Symbolic Multi-Step Reasoning Taskcited by 49
  36. 2024What does the Knowledge Neuron Thesis Have to do with Knowledge?cited by 49
  37. 2025MIB: A Mechanistic Interpretability Benchmarkcited by 49
  38. 2023Transformers Learn to Achieve Second-Order Convergence Rates for In-Context Linear Regressioncited by 47

50 most cited of 700 All 700 (JSON)