Pith. sign in

REVIEW 22 cited by

Towards Automated Circuit Discovery for Mechanistic Interpretability

Not yet reviewed by Pith; the record is open.

Re-run · record.json Download PDF Read on arXiv ↗

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2304.14997 v4 pith:5EI4AKGL submitted 2023-04-28 cs.LG

Arthur Conmy , Augustine N. Mavor-Parker , Aengus Lynch , Stefan Heimersheim , Adrià Garriga-Alonso This is my paper · ORCID
classification cs.LG
keywords behaviorcircuitinterpretabilityacdccomponentdatasetgpt-2mechanistic
verification ladder T0 review T1 audit T2 compute T3 formal

Signed reviews

No signed human review yet.

0 comments
read the original abstract

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.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 22 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. OpenAlex reports about 32 citations worldwide. Full citation record

  1. Where You Measure Decides What You Measure: Position Selection in Ablation-Based SAE Evaluation

    cs.LG 2026-08 accept novelty 7.0 of 10

    In ablation-based SAE evaluation, the measurement token is selected by the dictionary under test, and holding that token fixed collapses most of the variance that is usually attributed to differences between dictionaries.

  2. Reality Monitoring in Large Language Models: Self-Knowledge That Transforms with Conversation Memory

    cs.AI 2026-07 conditional novelty 7.0 of 10

    LLMs' source-attribution ability is not fixed: it flips with conversational memory structure, and corrective feedback can invert judgments or sever confidence from accuracy.

  3. LAWFUL: Law-Aligned Witness for Faithful Use of Latents

    cs.LG 2026-07 conditional novelty 7.0 of 10

    LAWFUL defines coverage-aware physical-consistency scores and circuit tests, reporting that a MoCap-to-Radar transformer's 9-component temporal circuit carries Doppler-law consistency via attention patterns.

  4. Sparse Weight Decomposition for Efficient Circuit Extraction

    cs.LG 2026-08 conditional novelty 6.0 of 10

    Sparse Weight Decomposition reparameterizes transformer weight matrices into sparse factors whose bottleneck units support efficient circuit extraction with less data and sparser circuits than learned sparse baselines.

  5. IFCLoRA: Topology-Aware Rank Allocation for Parameter-Efficient Fine-Tuning

    cs.LG 2026-07 conditional novelty 6.0 of 10

    IFCLoRA allocates LoRA ranks before fine-tuning using information-flow centrality from a calibration-set interaction graph, improving GSM8K by 1.36–1.82 points over LoRA at low ranks.

  6. Denoising Models Develop Human-Like Perceptual Illusion Representations Across Architectures

    cs.CV 2026-07 conditional novelty 6.0 of 10

    Denoising diffusion models encode visual illusions in internal layers, yet these representations do not influence the generated image.

  7. Targeted Recovery of Weight-Space Mechanisms From Neural Networks

    cs.LG 2026-06 conditional novelty 6.0 of 10

    A targeted decomposition method recovers the weight-space mechanisms behind specific inputs at low FLOPs, enabling focused ablation and rewiring of a 12-block transformer.

  8. Routing Subspaces: Auditing Evaluation-to-Deployment Mismatch in Fine-Tuned Language Models

    cs.CL 2026-05 conditional novelty 6.0 of 10

    Removing a single fitted activation direction at a mid-depth layer reduces the evaluation-vs-deployment behavioral gap on held-out prompts in 10 of 12 fine-tuned LLM settings, with matched controls staying flat.

  9. Language Model Circuits Are Sparse in the Neuron Basis

    cs.CL 2026-01 conditional novelty 6.0 of 10

    MLP neuron activations are shown to be as sparse and faithful a basis for circuit tracing as sparse autoencoder features, enabling simplified interpretability pipelines.

  10. Circuit Stability Characterizes Language Model Generalization

    cs.CL 2025-05 reject novelty 6.0 of 10

    Circuit stability, measured as rank correlation between soft circuits across subtasks, is proposed as a predictor of language model generalization.

  11. Beyond Induction Heads: In-Context Meta Learning Induces Multi-Phase Circuit Emergence

    cs.CL 2025-05 conditional novelty 6.0 of 10

    A two-layer transformer solving an in-context meta-learning task acquires skill in three abrupt phases, each corresponding to a distinct attention circuit: bigram, label attention, then chunking plus label attention.

  12. Prisma: An Open Source Toolkit for Mechanistic Interpretability in Vision and Video

    cs.CV 2025-04 conditional novelty 6.0 of 10

    Prisma is an open-source toolkit for vision/video mechanistic interpretability with 75+ hooked models and 80+ released sparse autoencoder weights, plus early evidence that vision SAEs are denser and can slightly reduce loss.

  13. Transcoders Beat Sparse Autoencoders for Interpretability

    cs.LG 2025-01 conditional novelty 6.0 of 10

    Skip transcoders beat sparse autoencoders on both reconstruction fidelity and automated interpretability scores for transformer MLP layers.

  14. MechELK: A Mechanistic Interpretability Framework for Eliciting Latent Knowledge in Large Language Models

    cs.CL 2026-04 conditional novelty 5.0 of 10

    MechELK combines SAE features, causal patching, and representation engineering to elicit latent LLM knowledge at 84.7% average accuracy, beating CCS by 6.2%.

  15. Enhancing Multi-Robot Exploration Using Probabilistic Frontier Prioritization with Dirichlet Process Gaussian Mixtures

    cs.RO 2026-04 unverdicted novelty 5.0 of 10

    DP-GMM-based probabilistic frontier prioritization improves two multi-agent frontier explorers by roughly 10–14% across clutter, team size, and communication settings.

  16. From Features to Actions: Explainability in Traditional and Agentic AI Systems

    cs.AI 2026-02 conditional novelty 5.0 of 10

    Attribution explanations that work for static classifiers do not diagnose failures in multi-step AI agents; trace-grounded rubric evaluation does, with state-tracking inconsistency 2.7x more common in failed agent runs.

  17. From Indirect Object Identification to Syllogisms: Exploring Binary Mechanisms in Transformer Circuits

    cs.CL 2025-08 conditional novelty 5.0 of 10

    GPT-2 small performs syllogisms through truth-copying attention heads and a suppression-plus-MLP pathway that can output a negated truth value.

  18. Evaluating Explanations: An Explanatory Virtues Framework for Mechanistic Interpretability -- The Strange Science Part I.ii

    cs.LG 2025-05 conditional novelty 5.0 of 10

    The paper introduces an Explanatory Virtues Framework and argues, via a qualitative rubric, that Compact Proofs are the most promising method for mechanistic interpretability.

  19. Towards Inclusive Toxic Content Moderation: Addressing Vulnerabilities to Adversarial Attacks in Toxicity Classifiers Tackling LLM-generated Content

    cs.CL 2025-09 reject novelty 4.0 of 10

    Zeroing attack-vulnerable attention heads improves BERT/RoBERTa toxicity classifier accuracy on PGD-adversarial inputs, with distinct heads implicated per demographic group.

  20. Perspectives for Direct Interpretability in Multi-Agent Deep Reinforcement Learning

    cs.AI 2025-02 unverdicted novelty 4.0 of 10

    A perspective paper that advocates direct, post hoc interpretability for multi-agent deep reinforcement learning and offers a taxonomy of where those methods might apply.

  21. AI Governance through Markets

    econ.GN 2025-01 conditional novelty 4.0 of 10

    Market governance mechanisms, supported by standardized AI disclosures, can create financial incentives for responsible AI development, according to this policy paper.

  22. Mechanistic Interpretability for Neural Networks: Circuits, Sparse Features and Symbolic Reasoning

    cs.LG 2026-07 accept

    A scoping review surveying circuit analysis, sparse autoencoders, activation steering, and neurosymbolic frameworks for interpreting and controlling Transformer-based neural networks.

Pith tools