{"total":156,"rows":[{"id":"2404.14082","title":"Mechanistic Interpretability for AI Safety -- A Review","created":"2024-04-22","cited_by_count":422,"area":"mechinterp"},{"id":"2309.16042","title":"Towards Best Practices of Activation Patching in Language Models: Metrics and Methods","created":"2023-09-27","cited_by_count":359,"area":"mechinterp"},{"id":"2309.01029","title":"Explainability for Large Language Models: A Survey","created":"2023-09-02","cited_by_count":333,"area":null},{"id":"2309.00941","title":"Emergent Linear Representations in World Models of Self-Supervised Sequence Models","created":"2023-09-02","cited_by_count":319,"area":"wm"},{"id":"2404.15255","title":"How to use and interpret activation patching","created":"2024-04-23","cited_by_count":214,"area":null},{"id":"2402.16837","title":"Do Large Language Models Latently Perform Multi-Hop Reasoning?","created":"2024-02-26","cited_by_count":169,"area":"lr"},{"id":"2406.19384","title":"The Remarkable Robustness of LLMs: Stages of Inference?","created":"2024-06-27","cited_by_count":146,"area":null},{"id":"2403.00745","title":"AtP*: An efficient and scalable method for localizing LLM behaviour to components","created":"2024-03-01","cited_by_count":111,"area":"mechinterp"},{"id":"2401.12181","title":"Universal Neurons in GPT2 Language Models","created":"2024-01-22","cited_by_count":102,"area":"mechinterp"},{"id":"2403.00824","title":"Information Flow Routes: Automatically Interpreting Language Models at Scale","created":"2024-02-27","cited_by_count":95,"area":null},{"id":"2405.08366","title":"Towards Principled Evaluations of Sparse Autoencoders for Interpretability and Control","created":"2024-05-14","cited_by_count":86,"area":"mechinterp"},{"id":"2405.00208","title":"A Primer on the Inner Workings of Transformer-based Language Models","created":"2024-04-30","cited_by_count":85,"area":null},{"id":"2406.12775","title":"Hopping Too Late: Exploring the Limitations of Large Language Models on Multi-Hop Queries","created":"2024-06-18","cited_by_count":78,"area":null},{"id":"2502.15603","title":"Do Multilingual LLMs Think In English?","created":"2025-02-21","cited_by_count":78,"area":"lr"},{"id":"2312.09230","title":"Successor Heads: Recurring, Interpretable Attention Heads In The Wild","created":"2023-12-14","cited_by_count":77,"area":null},{"id":"2310.04625","title":"Copy Suppression: Comprehensively Understanding an Attention Head","created":"2023-10-06","cited_by_count":76,"area":null},{"id":"2407.10827","title":"LLM Circuit Analyses Are Consistent Across Training and Scale","created":"2024-07-15","cited_by_count":63,"area":null},{"id":"2406.16778","title":"Finding Transformer Circuits with Edge Pruning","created":"2024-06-24","cited_by_count":60,"area":null},{"id":"2406.01563","title":"LoFiT: Localized Fine-tuning on LLM Representations","created":"2024-06-03","cited_by_count":57,"area":null},{"id":"2402.18312","title":"How to think step-by-step: A mechanistic understanding of chain-of-thought reasoning","created":"2024-02-28","cited_by_count":53,"area":"lr"},{"id":"2406.14144","title":"Towards Understanding Safety Alignment: A Mechanistic Perspective from Safety Neurons","created":"2024-06-20","cited_by_count":53,"area":"alignment"},{"id":"2510.26243","title":"Angular Steering: Behavior Control via Rotation in Activation Space","created":"2025-10-30","cited_by_count":53,"area":null},{"id":"2402.05406","title":"Everybody Prune Now: Structured Pruning of LLMs with only Forward Passes","created":"2024-02-08","cited_by_count":43,"area":null},{"id":"2402.07321","title":"Summing Up the Facts: Additive Mechanisms Behind Factual Recall in LLMs","created":"2024-02-11","cited_by_count":42,"area":null},{"id":"2403.01267","title":"Dissecting Language Models: Machine Unlearning via Selective Pruning","created":"2024-03-02","cited_by_count":42,"area":"safety"},{"id":"2402.15390","title":"Explorations of Self-Repair in Language Models","created":"2024-02-23","cited_by_count":39,"area":null},{"id":"2406.04341","title":"Interpreting the Second-Order Effects of Neurons in CLIP","created":"2024-06-06","cited_by_count":36,"area":null},{"id":"2410.04332","title":"Gradient Routing: Masking Gradients to Localize Computation in Neural Networks","created":"2024-10-06","cited_by_count":34,"area":null},{"id":"2403.19521","title":"Interpreting Key Mechanisms of Factual Recall in Transformer-Based Language Models","created":"2024-03-28","cited_by_count":31,"area":null},{"id":"2402.12201","title":"Dictionary Learning Improves Patch-Free Circuit Discovery in Mechanistic Interpretability: A Case Study on Othello-GPT","created":"2024-02-19","cited_by_count":29,"area":"mechinterp"},{"id":"2409.09951","title":"Optimal ablation for interpretability","created":"2024-09-16","cited_by_count":25,"area":null},{"id":"2505.24535","title":"Beyond Linear Steering: Unified Multi-Attribute Control for Language Models","created":"2025-05-30","cited_by_count":25,"area":null},{"id":"2312.06581","title":"Grokking Group Multiplication with Cosets","created":"2023-12-11","cited_by_count":24,"area":null},{"id":"2405.13868","title":"Automatically Identifying Local and Global Circuits with Linear Computation Graphs","created":"2024-05-22","cited_by_count":24,"area":"mechinterp"},{"id":"2408.01416","title":"The Quest for the Right Mediator: Surveying Mechanistic Interpretability Through the Lens of Causal Mediation Analysis","created":"2024-08-02","cited_by_count":24,"area":"mechinterp"},{"id":"2407.04690","title":"Missed Causes and Ambiguous Effects: Counterfactuals Pose Challenges for Interpreting Neural Networks","created":"2024-07-05","cited_by_count":23,"area":null},{"id":"2310.16270","title":"Attention Lens: A Tool for Mechanistically Interpreting the Attention Head Information Retrieval Mechanism","created":"2023-10-25","cited_by_count":22,"area":null},{"id":"2401.03735","title":"Language Models Encode the Value of Numbers Linearly","created":"2024-01-08","cited_by_count":20,"area":"mechinterp"},{"id":"2509.09708","title":"Beyond I'm Sorry, I Can't: Dissecting Large Language Model Refusal","created":"2025-09-07","cited_by_count":19,"area":"safety"},{"id":"2312.10091","title":"Look Before You Leap: A Universal Emergent Decomposition of Retrieval Tasks in Language Models","created":"2023-12-13","cited_by_count":18,"area":null},{"id":"2408.03247","title":"Unveiling Factual Recall Behaviors of Large Language Models through Knowledge Neurons","created":"2024-08-06","cited_by_count":16,"area":null},{"id":"2506.20790","title":"Stochastic Parameter Decomposition","created":"2025-06-25","cited_by_count":16,"area":null},{"id":"2403.10381","title":"Monotonic Representation of Numeric Properties in Language Models","created":"2024-03-15","cited_by_count":15,"area":null},{"id":"2406.18400","title":"Do LLMs dream of elephants (when told not to)? Latent concept association and associative memory in transformers","created":"2024-06-26","cited_by_count":15,"area":null},{"id":"2411.04105","title":"A Implies B: Circuit Analysis in LLMs for Propositional Logical Reasoning","created":"2024-11-06","cited_by_count":15,"area":"lr"},{"id":"2503.07604","title":"Implicit Reasoning in Transformers is Reasoning through Shortcuts","created":"2025-03-10","cited_by_count":14,"area":"lr"},{"id":"2407.19842","title":"Detecting and Understanding Vulnerabilities in Language Models via Mechanistic Interpretability","created":"2024-07-29","cited_by_count":13,"area":"mechinterp"},{"id":"2502.03032","title":"Analyze Feature Flow to Enhance Interpretation and Steering in Language Models","created":"2025-02-05","cited_by_count":13,"area":"mechinterp"},{"id":"2512.18792","title":"The Dead Salmons of AI Interpretability","created":"2025-12-21","cited_by_count":12,"area":null},{"id":"2512.24440","title":"Towards mechanistic understanding in a data-driven weather model: internal activations reveal interpretable physical features","created":"2025-12-30","cited_by_count":12,"area":null},{"id":"2310.07325","title":"An Adversarial Example for Direct Logit Attribution: Memory Management in GELU-4L","created":"2023-10-11","cited_by_count":11,"area":null},{"id":"2407.06677","title":"Mixture-of-Modules: Reinventing Transformers as Dynamic Assemblies of Modules","created":"2024-07-09","cited_by_count":11,"area":null},{"id":"2503.11302","title":"Are formal and functional linguistic mechanisms dissociated in language models?","created":"2025-03-14","cited_by_count":11,"area":null},{"id":"2511.07979","title":"Benchmarking Multi-Step Legal Reasoning and Analyzing Chain-of-Thought Effects in Large Language Models","created":"2025-11-11","cited_by_count":10,"area":"lr"},{"id":"2604.19117","title":"LLMs Know They're Wrong and Agree Anyway: The Shared Sycophancy-Lying Circuit","created":"2026-04-21","cited_by_count":10,"area":null},{"id":"2605.06510","title":"Is One Layer Enough? Understanding Inference Dynamics in Tabular Foundation Models","created":"2026-05-07","cited_by_count":10,"area":null},{"id":"2509.17588","title":"Interpreting Attention Heads for Image-to-Text Information Flow in Large Vision-Language Models","created":"2025-09-22","cited_by_count":9,"area":"mllm"},{"id":"2511.21399","title":"Steering Awareness: Detecting Activation Steering from Within","created":"2025-11-26","cited_by_count":9,"area":"mechinterp"},{"id":"2605.12671","title":"All Circuits Lead to Rome: Rethinking Functional Anisotropy in Circuit and Sheaf Discovery for LLMs","created":"2026-05-12","cited_by_count":9,"area":"lr"},{"id":"2607.01940","title":"Conditional Co-Ablation: Recovering Self-Repair Backups in Transformer Circuits","created":"2026-07-02","cited_by_count":9,"area":"mechinterp"},{"id":"2601.22364","title":"Context Structure Reshapes the Representational Geometry of Language Models","created":"2026-01-29","cited_by_count":8,"area":null},{"id":"2602.16698","title":"Causality is Key for Interpretability Claims to Generalise","created":"2026-02-18","cited_by_count":8,"area":null},{"id":"2602.22424","title":"Causality $\\neq$ Invariance: Function and Concept Vectors in LLMs","created":"2026-02-25","cited_by_count":8,"area":null},{"id":"2311.00863","title":"Training Dynamics of Contextual N-Grams in Language Models","created":"2023-11-01","cited_by_count":7,"area":null},{"id":"2312.16374","title":"LLM Factoscope: Uncovering LLMs' Factual Discernment through Inner States Analysis","created":"2023-12-27","cited_by_count":7,"area":null},{"id":"2406.16990","title":"AND: Audio Network Dissection for Interpreting Deep Acoustic Models","created":"2024-06-24","cited_by_count":7,"area":null},{"id":"2504.07898","title":"How do Large Language Models Understand Relevance? A Mechanistic Interpretability Perspective","created":"2025-04-10","cited_by_count":7,"area":"mechinterp"},{"id":"2510.26784","title":"LLMs Process Lists With General Filter Heads","created":"2025-10-30","cited_by_count":7,"area":null},{"id":"2310.02258","title":"A Neural Scaling Law from Lottery Ticket Ensembling","created":"2023-10-03","cited_by_count":5,"area":null},{"id":"2404.01903","title":"Understanding How CodeLLMs (Mis)Predict Types with Activation Steering","created":"2024-04-02","cited_by_count":5,"area":"mechinterp"},{"id":"2408.12664","title":"Multilevel Interpretability Of Artificial Neural Networks: Leveraging Framework And Methods From Neuroscience","created":"2024-08-22","cited_by_count":5,"area":null},{"id":"2409.17113","title":"Characterizing stable regions in the residual stream of LLMs","created":"2024-09-25","cited_by_count":5,"area":null},{"id":"2505.06032","title":"Short-circuiting Shortcuts: Mechanistic Investigation of Shortcuts in Text Classification","created":"2025-05-09","cited_by_count":5,"area":null},{"id":"2505.17630","title":"Correcting Gradient-Based Circuit Localization via Interaction-Aware Backpropagation","created":"2025-05-23","cited_by_count":5,"area":null},{"id":"2508.14067","title":"Punctuation and Predicates in Language Models","created":"2025-08-11","cited_by_count":5,"area":null},{"id":"2602.01605","title":"Universal Redundancies in Time Series Foundation Models","created":"2026-02-02","cited_by_count":5,"area":null},{"id":"2602.06941","title":"Endogenous Resistance to Activation Steering in Language Models","created":"2026-02-06","cited_by_count":5,"area":"mechinterp"},{"id":"2607.03358","title":"Pathways of Visual Information Flow in Vision-Language Models","created":"2026-07-03","cited_by_count":5,"area":"mllm"},{"id":"2410.00340","title":"Sparse Attention Decomposition Applied to Circuit Tracing","created":"2024-10-01","cited_by_count":4,"area":null},{"id":"2502.17355","title":"On Relation-Specific Neurons in Large Language Models","created":"2025-02-24","cited_by_count":4,"area":null},{"id":"2504.10077","title":"Towards Quantifying Commonsense Reasoning with Mechanistic Insights","created":"2025-04-14","cited_by_count":4,"area":"lr"},{"id":"2602.23164","title":"MetaOthello: A Controlled Study of Multiple World Models in Transformers","created":"2026-02-26","cited_by_count":4,"area":"wm"},{"id":"2606.12689","title":"Observable Patterns Are Not Explanations: A Causal-Geometric Analysis of Latent Reasoning Models","created":"2026-06-10","cited_by_count":4,"area":"lr"},{"id":"2312.05361","title":"Emergence and Function of Abstract Representations in Self-Supervised Transformers","created":"2023-12-08","cited_by_count":3,"area":"wm"},{"id":"2405.17799","title":"Exploring Activation Patterns of Parameters in Language Models","created":"2024-05-28","cited_by_count":3,"area":null},{"id":"2412.15750","title":"Extracting Interpretable Task-Specific Circuits from Large Language Models for Faster Inference","created":"2024-12-20","cited_by_count":3,"area":"mechinterp"},{"id":"2603.07461","title":"The Dual-Stream Transformer: Channelized Architecture for Interpretable Language Modeling","created":"2026-03-08","cited_by_count":3,"area":null},{"id":"2606.06267","title":"Many Circuits, One Mechanism: Input Variation and Evaluation Granularity in Circuit Discovery","created":"2026-06-04","cited_by_count":3,"area":"mechinterp"},{"id":"2608.10986","title":"What Iterated Self-Feeding Probes of Language Models Measure, and a test that separates the construction from the model","created":"2026-08-11","cited_by_count":3,"area":"lr"},{"id":"2609.07876","title":"LLM Layers Immediately Correct Each Other","created":"2026-09-07","cited_by_count":3,"area":"mechinterp"},{"id":"2504.13763","title":"Decoding Vision Transformers: the Diffusion Steering Lens","created":"2025-04-18","cited_by_count":2,"area":"mechinterp"},{"id":"2603.06592","title":"Hierarchical Latent Structures in Data Generation Process Unify Mechanistic Phenomena across Scale","created":"2026-02-04","cited_by_count":2,"area":"mechinterp"},{"id":"2604.10158","title":"Tracing the Thought of a Grandmaster-level Chess-Playing Transformer","created":"2026-04-11","cited_by_count":2,"area":null},{"id":"2605.09314","title":"How LLMs Are Persuaded: A Few Attention Heads, Rerouted","created":"2026-05-10","cited_by_count":2,"area":null},{"id":"2605.28839","title":"One Mask to Rule Them All: On Hidden Facts after Editing and How to Find Them","created":"2026-04-18","cited_by_count":2,"area":null},{"id":"2606.00033","title":"Make Mechanistic Interpretability Auditable: A Call to Develop Guidelines via Continuous Collaborative Reviewing","created":"2026-04-24","cited_by_count":2,"area":"mechinterp"},{"id":"2606.07560","title":"Function-Vector Heads Are Two Populations: Writers and Cancellers in In-Context Learning","created":"2026-05-25","cited_by_count":2,"area":null},{"id":"2607.17568","title":"CoCurve: Cross-Module Co-Pruning Curvature for Structured LLM Pruning","created":"2026-07-20","cited_by_count":2,"area":null},{"id":"2608.19338","title":"Mechanistic Tomography: Designed Measurement for Control-Oriented Interpretability","created":"2026-08-19","cited_by_count":2,"area":"mechinterp"},{"id":"2403.08739","title":"The Garden of Forking Paths: Observing Dynamic Parameters Distribution in Large Language Models","created":"2024-03-13","cited_by_count":1,"area":null}],"next_cursor":"100"}