{"total":359,"rows":[{"id":"2404.14082","title":"Mechanistic Interpretability for AI Safety -- A Review","created":"2024-04-22","cited_by_count":422,"area":"mechinterp"},{"id":"2401.06102","title":"Patchscopes: A Unifying Framework for Inspecting Hidden Representations of Language Models","created":"2024-01-11","cited_by_count":235,"area":"mechinterp"},{"id":"2404.15255","title":"How to use and interpret activation patching","created":"2024-04-23","cited_by_count":214,"area":null},{"id":"2405.14860","title":"Not All Language Model Features Are One-Dimensionally Linear","created":"2024-05-23","cited_by_count":212,"area":null},{"id":"2406.11944","title":"Transcoders Find Interpretable LLM Feature Circuits","created":"2024-06-17","cited_by_count":200,"area":"mechinterp"},{"id":"2403.17806","title":"Have Faith in Faithfulness: Going Beyond Circuit Overlap When Finding Model Mechanisms","created":"2024-03-26","cited_by_count":149,"area":null},{"id":"2406.19384","title":"The Remarkable Robustness of LLMs: Stages of Inference?","created":"2024-06-27","cited_by_count":146,"area":null},{"id":"2407.02646","title":"A Practical Review of Mechanistic Interpretability for Transformer-Based Language Models","created":"2024-07-02","cited_by_count":118,"area":"mechinterp"},{"id":"2402.18154","title":"Cutting Off the Head Ends the Conflict: A Mechanism for Interpreting and Mitigating Knowledge Conflicts in Language Models","created":"2024-02-28","cited_by_count":91,"area":null},{"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":"2405.15071","title":"Grokked Transformers are Implicit Reasoners: A Mechanistic Journey to the Edge of Generalization","created":"2024-05-23","cited_by_count":79,"area":"lr"},{"id":"2310.14735","title":"Unleashing the potential of prompt engineering for large language models","created":"2023-10-23","cited_by_count":72,"area":null},{"id":"2410.13708","title":"On the Role of Attention Heads in Large Language Model Safety","created":"2024-10-17","cited_by_count":72,"area":"safety"},{"id":"2409.03752","title":"Attention Heads of Large Language Models: A Survey","created":"2024-09-05","cited_by_count":67,"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":"2504.13151","title":"MIB: A Mechanistic Interpretability Benchmark","created":"2025-04-17","cited_by_count":56,"area":"mechinterp"},{"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":"2407.15017","title":"Knowledge Mechanisms in Large Language Models: A Survey and Perspective","created":"2024-07-22","cited_by_count":51,"area":null},{"id":"2410.05864","title":"From Tokens to Words: On the Inner Lexicon of LLMs","created":"2024-10-08","cited_by_count":48,"area":null},{"id":"2407.14561","title":"NNsight and NDIF: Democratizing Access to Open-Weight Foundation Model Internals","created":"2024-07-18","cited_by_count":45,"area":null},{"id":"2403.20009","title":"On Large Language Models' Hallucination with Regard to Known Facts","created":"2024-03-29","cited_by_count":43,"area":null},{"id":"2505.13763","title":"Language Models Are Capable of Metacognitive Monitoring and Control of Their Internal Activations","created":"2025-05-19","cited_by_count":41,"area":null},{"id":"2312.12141","title":"Neuron-Level Knowledge Attribution in Large Language Models","created":"2023-12-19","cited_by_count":37,"area":null},{"id":"2404.03646","title":"Locating and Editing Factual Associations in Mamba","created":"2024-04-04","cited_by_count":37,"area":"ssm"},{"id":"2502.19249","title":"Between Circuits and Chomsky: Pre-pretraining on Formal Languages Imparts Linguistic Biases","created":"2025-02-26","cited_by_count":37,"area":null},{"id":"2407.15018","title":"Answer, Assemble, Ace: Understanding How LMs Answer Multiple Choice Questions","created":"2024-07-21","cited_by_count":35,"area":null},{"id":"2406.00877","title":"Evidence of Learned Look-Ahead in a Chess-Playing Neural Network","created":"2024-06-02","cited_by_count":34,"area":null},{"id":"2503.02854","title":"(How) Do Language Models Track State?","created":"2025-03-04","cited_by_count":34,"area":null},{"id":"2411.04430","title":"Towards Unifying Interpretability and Control: Evaluation via Intervention","created":"2024-11-07","cited_by_count":32,"area":"mechinterp"},{"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":"2403.19851","title":"Localizing Paragraph Memorization in Language Models","created":"2024-03-28","cited_by_count":31,"area":null},{"id":"2403.05846","title":"Diffusion Lens: Interpreting Text Encoders in Text-to-Image Pipelines","created":"2024-03-09","cited_by_count":30,"area":"diffusion"},{"id":"2406.16320","title":"What Do VLMs NOTICE? A Mechanistic Interpretability Pipeline for Gaussian-Noise-free Text-Image Corruption and Evaluation","created":"2024-06-24","cited_by_count":30,"area":"mllm"},{"id":"2511.08579","title":"Training Language Models to Explain Their Own Computations","created":"2025-11-11","cited_by_count":30,"area":null},{"id":"2402.09236","title":"Learning Interpretable Concepts: Unifying Causal Representation Learning and Foundation Models","created":"2024-02-14","cited_by_count":29,"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":"2405.16229","title":"No Two Devils Alike: Unveiling Distinct Mechanisms of Fine-tuning Attacks","created":"2024-05-25","cited_by_count":29,"area":"alignment"},{"id":"2410.13032","title":"Hypothesis Testing the Circuit Hypothesis in LLMs","created":"2024-10-16","cited_by_count":29,"area":"mechinterp"},{"id":"2601.14004","title":"Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models","created":"2026-01-20","cited_by_count":27,"area":"mechinterp"},{"id":"2603.17839","title":"How do LLMs Compute Verbal Confidence","created":"2026-03-18","cited_by_count":26,"area":null},{"id":"2407.14494","title":"InterpBench: Semi-Synthetic Transformers for Evaluating Mechanistic Interpretability Techniques","created":"2024-07-19","cited_by_count":25,"area":"mechinterp"},{"id":"2409.09951","title":"Optimal ablation for interpretability","created":"2024-09-16","cited_by_count":25,"area":null},{"id":"2404.05729","title":"Finding Visual Task Vectors","created":"2024-04-08","cited_by_count":24,"area":null},{"id":"2604.05655","title":"LLM Reasoning as Trajectories: Step-Specific Representation Geometry and Correctness Signals","created":"2026-04-07","cited_by_count":24,"area":"lr"},{"id":"2406.12274","title":"SafeInfer: Context Adaptive Decoding Time Safety Alignment for Large Language Models","created":"2024-06-18","cited_by_count":23,"area":"alignment"},{"id":"2502.10835","title":"Back Attention: Understanding and Enhancing Multi-Hop Reasoning in Large Language Models","created":"2025-02-15","cited_by_count":23,"area":"lr"},{"id":"2407.08734","title":"Transformer Circuit Faithfulness Metrics are not Robust","created":"2024-07-11","cited_by_count":22,"area":"mechinterp"},{"id":"2411.11114","title":"JailbreakLens: Interpreting Jailbreak Mechanism in the Lens of Representation and Circuit","created":"2024-11-17","cited_by_count":22,"area":"safety"},{"id":"2502.17601","title":"Representation Engineering for Large-Language Models: Survey and Research Challenges","created":"2025-02-24","cited_by_count":21,"area":"mechinterp"},{"id":"2404.11534","title":"Decomposing and Editing Predictions by Modeling Model Computation","created":"2024-04-17","cited_by_count":19,"area":null},{"id":"2412.04703","title":"Transformers Struggle to Learn to Search","created":"2024-12-06","cited_by_count":19,"area":null},{"id":"2507.22928","title":"How does Chain of Thought Think? Mechanistic Interpretability of Chain-of-Thought Reasoning with Sparse Autoencoding","created":"2025-07-24","cited_by_count":19,"area":"lr"},{"id":"2405.03207","title":"A Philosophical Introduction to Language Models - Part II: The Way Forward","created":"2024-05-06","cited_by_count":18,"area":null},{"id":"2405.17345","title":"Exploring and steering the moral compass of Large Language Models","created":"2024-05-27","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":"2405.12522","title":"Sparse Autoencoders Enable Scalable and Reliable Circuit Identification in Language Models","created":"2024-05-21","cited_by_count":15,"area":"mechinterp"},{"id":"2410.22330","title":"Vision-Language Models Create Cross-Modal Task Representations","created":"2024-10-29","cited_by_count":15,"area":"mllm"},{"id":"2506.00253","title":"Aligned but Blind: Alignment Increases Implicit Bias by Reducing Awareness of Race","created":"2025-05-30","cited_by_count":15,"area":"alignment"},{"id":"2506.05166","title":"Dissecting Bias in LLMs: A Mechanistic Interpretability Perspective","created":"2025-06-05","cited_by_count":15,"area":"mechinterp"},{"id":"2507.07417","title":"May I have your Attention? Breaking Fine-Tuning based Prompt Injection Defenses using Architecture-Aware Attacks","created":"2025-07-10","cited_by_count":15,"area":"safety"},{"id":"2510.00845","title":"Mechanistic Interpretability as Statistical Estimation: A Variance Analysis","created":"2025-10-01","cited_by_count":15,"area":"mechinterp"},{"id":"2402.16726","title":"Towards Empirical Interpretation of Internal Circuits and Properties in Grokked Transformers on Modular Polynomials","created":"2024-02-26","cited_by_count":14,"area":null},{"id":"2407.03779","title":"Sheaf Discovery with Joint Computation Graph Pruning and Flexible Granularity","created":"2024-07-04","cited_by_count":14,"area":null},{"id":"2410.14155","title":"Towards Faithful Natural Language Explanations: A Study Using Activation Patching in Large Language Models","created":"2024-10-18","cited_by_count":14,"area":"mechinterp"},{"id":"2503.07604","title":"Implicit Reasoning in Transformers is Reasoning through Shortcuts","created":"2025-03-10","cited_by_count":14,"area":"lr"},{"id":"2601.12626","title":"Linear Mechanisms for Spatiotemporal Reasoning in Vision Language Models","created":"2026-01-18","cited_by_count":14,"area":"lr"},{"id":"2501.06751","title":"Padding Tone: A Mechanistic Analysis of Padding Tokens in T2I Models","created":"2025-01-12","cited_by_count":13,"area":"diffusion"},{"id":"2505.17697","title":"Activation Control for Efficiently Eliciting Long Chain-of-thought Ability of Language Models","created":"2025-05-23","cited_by_count":13,"area":"lr"},{"id":"2512.18792","title":"The Dead Salmons of AI Interpretability","created":"2025-12-21","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":"2408.15510","title":"How Reliable are Causal Probing Interventions?","created":"2024-08-28","cited_by_count":11,"area":null},{"id":"2410.01434","title":"Circuit Compositions: Exploring Modular Structures in Transformer-Based Language Models","created":"2024-10-02","cited_by_count":11,"area":null},{"id":"2410.22244","title":"Abrupt Learning in Transformers: A Case Study on Matrix Completion","created":"2024-10-29","cited_by_count":11,"area":null},{"id":"2411.10950","title":"Understanding Multimodal LLMs: the Mechanistic Interpretability of Llava in Visual Question Answering","created":"2024-11-17","cited_by_count":11,"area":"mllm"},{"id":"2412.01113","title":"LLMs Faithfully and Iteratively Compute Answers During CoT: A Systematic Analysis With Multi-step Arithmetics","created":"2024-12-02","cited_by_count":11,"area":"lr"},{"id":"2412.02542","title":"Unveiling Concept Attribution in Diffusion Models","created":"2024-12-03","cited_by_count":11,"area":"diffusion"},{"id":"2508.11214","title":"How Causal Abstraction Underpins Computational Explanation","created":"2025-08-15","cited_by_count":11,"area":null},{"id":"2510.06036","title":"Refusal Falls off a Cliff: How Safety Alignment Fails in Reasoning?","created":"2025-10-07","cited_by_count":11,"area":"lr"},{"id":"2512.12069","title":"Rethinking Jailbreak Detection of Large Vision Language Models with Representational Contrastive Scoring","created":"2025-12-12","cited_by_count":11,"area":"lr"},{"id":"2602.16823","title":"Formal Mechanistic Interpretability: Automated Circuit Discovery with Provable Guarantees","created":"2026-02-18","cited_by_count":11,"area":"mechinterp"},{"id":"2602.17053","title":"RFEval: Benchmarking Reasoning Faithfulness under Counterfactual Reasoning Intervention in Large Reasoning Models","created":"2026-02-19","cited_by_count":11,"area":"lr"},{"id":"2403.03942","title":"The Heuristic Core: Understanding Subnetwork Generalization in Pretrained Language Models","created":"2024-03-06","cited_by_count":10,"area":null},{"id":"2502.04667","title":"Compositional Generalization from Learned Skills via CoT Training: A Theoretical and Structural Analysis for Reasoning","created":"2025-02-07","cited_by_count":10,"area":"lr"},{"id":"2502.11771","title":"The Validation Gap: A Mechanistic Analysis of How Language Models Compute Arithmetic but Fail to Validate It","created":"2025-02-17","cited_by_count":10,"area":null},{"id":"2502.18147","title":"Jacobian Sparse Autoencoders: Sparsify Computations, Not Just Activations","created":"2025-02-25","cited_by_count":10,"area":"mechinterp"},{"id":"2505.10039","title":"Rethinking Circuit Completeness in Language Models: AND, OR, and ADDER Gates","created":"2025-05-15","cited_by_count":10,"area":"lr"},{"id":"2506.12880","title":"Universal Jailbreak Suffixes Are Strong Attention Hijackers","created":"2025-06-15","cited_by_count":10,"area":"safety"},{"id":"2409.06328","title":"Extracting Paragraphs from LLM Token Activations","created":"2024-09-10","cited_by_count":9,"area":null},{"id":"2504.20938","title":"Towards Understanding the Nature of Attention with Low-Rank Sparse Decomposition","created":"2025-04-29","cited_by_count":9,"area":"mechinterp"},{"id":"2505.13541","title":"SPIRIT: Patching Speech Language Models against Jailbreak Attacks","created":"2025-05-18","cited_by_count":9,"area":"safety"},{"id":"2604.02605","title":"Do Audio-Visual Large Language Models Really See and Hear?","created":"2026-04-03","cited_by_count":9,"area":null},{"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":"2408.14915","title":"Can Transformers Do Enumerative Geometry?","created":"2024-08-27","cited_by_count":8,"area":null},{"id":"2502.01179","title":"Joint Localization and Activation Editing for Low-Resource Fine-Tuning","created":"2025-02-03","cited_by_count":8,"area":null},{"id":"2505.23701","title":"Can LLMs Reason Abstractly Over Math Word Problems Without CoT? Disentangling Abstract Formulation From Arithmetic Computation","created":"2025-05-29","cited_by_count":8,"area":"lr"},{"id":"2507.06445","title":"Can Interpretation Predict Behavior on Unseen Data?","created":"2025-07-08","cited_by_count":8,"area":null},{"id":"2509.14837","title":"V-SEAM: Visual Semantic Editing and Attention Modulating for Causal Interpretability of Vision-Language Models","created":"2025-09-18","cited_by_count":8,"area":"mllm"},{"id":"2602.16698","title":"Causality is Key for Interpretability Claims to Generalise","created":"2026-02-18","cited_by_count":8,"area":null},{"id":"2602.20330","title":"Circuit Tracing in Vision-Language Models: Understanding the Internal Mechanisms of Multimodal Thinking","created":"2026-02-23","cited_by_count":8,"area":"lr"}],"next_cursor":"100"}