Path patching provides a method to express and quantitatively test hypotheses that neural network behaviors are localized to sets of paths.
hub
Causal abstraction for faithful model interpretation
17 Pith papers cite this work, alongside 10 external citations. Polarity classification is still indexing.
17
Pith papers citing it
10
external citations · external index
hub tools
citation-role summary
background 1
citation-polarity summary
roles
background 1polarities
background 1representative citing papers
Under consequence-invariant posterior training and sparsity of coordinated harm patterns, the training mass on dangerous guarded Predictors is bounded by C_bad times R_shell.
EPB distills NCO models into evolving program portfolios via LLM-driven textual-numerical optimization, matching original performance while exposing stage-dependent heuristic-like behavior.
LLMs show a probe-distance sign flip between finite and infinitival embedded clauses that UD cannot explain, claimed as evidence for Minimalist phase structure.
The paper introduces compositional interpretability as a category-theoretic framework that casts mechanistic explanations as commuting syntactic-semantic mappings optimized under faithfulness and complexity constraints derived from minimum description length.
A four-step recipe partitions the input space using interchange intervention behavior to diagnose where causal abstractions hold and to guide improvements, demonstrated by recovering a full hypothesis from scratch in a toy logic task.
Temporal preference in Qwen3-4B-Instruct-2507 localizes to layers 17–35 (especially L24 attention), has curved residual-stream geometry, is behaviorally unstable, and can be bidirectionally steered.
JSAE jointly factorizes pooled vision and language activations in VLMs into aligned interpretable features, revealing layer-dependent asymmetry in additive steering versus suppression on three models.
XtrAIn shifts occlusion from input space to parameter space along the training trajectory to produce cleaner feature attributions than standard methods.
XWP and XWP_c are novel attribution methods for FCNNs that estimate feature importance by perturbing attached weights to avoid added bias and out-of-distribution issues in occlusion approaches.
A latent mediation framework with sparse autoencoders enables non-additive token-level influence attribution in LLMs by learning orthogonal features and back-propagating attributions.
Sparse feature circuits are introduced as interpretable causal subnetworks in language models, supporting unsupervised discovery of thousands of circuits and a method called SHIFT to improve classifier generalization by ablating irrelevant features.
Sentiment is represented as a single linear direction in LLM activation space that is causally relevant across tasks and is summarized at punctuation and names in addition to charged words.
Reasoning in large output spaces proceeds via shortlisting then fine-grained reasoning; this characterization enables a mechanistic distillation strategy that outperforms standard distillation.
LLMs encode causal direction internally via probes but revert to commonsense in Yes/No outputs on anti-commonsense items, showing output accuracy alone does not measure causal understanding.
LLMs support decision prediction and rationale generation but lack evidence for genuine decision explanation, requiring stricter standards to avoid over-crediting.
citing papers explorer
-
Localizing Model Behavior with Path Patching
Path patching provides a method to express and quantitatively test hypotheses that neural network behaviors are localized to sets of paths.
-
Safety from Honesty in a Disinterested AI Predictor
Under consequence-invariant posterior training and sparsity of coordinated harm patterns, the training mass on dangerous guarded Predictors is bounded by C_bad times R_shell.
-
Interpreting Neural Combinatorial Optimization via Evolving Programmatic Bottlenecks
EPB distills NCO models into evolving program portfolios via LLM-driven textual-numerical optimization, matching original performance while exposing stage-dependent heuristic-like behavior.
-
Probing LLMs for Syntactic Structure Beyond Universal Dependencies: A Minimalist Phase Account in English
LLMs show a probe-distance sign flip between finite and infinitival embedded clauses that UD cannot explain, claimed as evidence for Minimalist phase structure.
-
From Mechanistic to Compositional Interpretability
The paper introduces compositional interpretability as a category-theoretic framework that casts mechanistic explanations as commuting syntactic-semantic mappings optimized under faithfulness and complexity constraints derived from minimum description length.
-
Bucketing the Good Apples: A Method for Diagnosing and Improving Causal Abstraction
A four-step recipe partitions the input space using interchange intervention behavior to diagnose where causal abstractions hold and to guide improvements, demonstrated by recovering a full hypothesis from scratch in a toy logic task.
-
Temporal Preference Concepts and their Functions in a Large Language Model
Temporal preference in Qwen3-4B-Instruct-2507 localizes to layers 17–35 (especially L24 attention), has curved residual-stream geometry, is behaviorally unstable, and can be bidirectionally steered.
-
Steering Vision-Language Models with Joint Sparse Autoencoders
JSAE jointly factorizes pooled vision and language activations in VLMs into aligned interpretable features, revealing layer-dependent asymmetry in additive steering versus suppression on three models.
-
XtrAIn: Training-Guided Occlusion for Feature Attribution
XtrAIn shifts occlusion from input space to parameter space along the training trajectory to produce cleaner feature attributions than standard methods.
-
From Weight Perturbation to Feature Attribution for Explaining Fully Connected Neural Networks
XWP and XWP_c are novel attribution methods for FCNNs that estimate feature importance by perturbing attached weights to avoid added bias and out-of-distribution issues in occlusion approaches.
-
Correcting Influence: Unboxing LLM Outputs with Orthogonal Latent Spaces
A latent mediation framework with sparse autoencoders enables non-additive token-level influence attribution in LLMs by learning orthogonal features and back-propagating attributions.
-
Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models
Sparse feature circuits are introduced as interpretable causal subnetworks in language models, supporting unsupervised discovery of thousands of circuits and a method called SHIFT to improve classifier generalization by ablating irrelevant features.
-
Linear Representations of Sentiment in Large Language Models
Sentiment is represented as a single linear direction in LLM activation space that is causally relevant across tasks and is summarized at punctuation and names in addition to charged words.
-
Characterize Then Distill: Mechanistic Reasoning in Large Output Spaces
Reasoning in large output spaces proceeds via shortlisting then fine-grained reasoning; this characterization enables a mechanistic distillation strategy that outperforms standard distillation.
-
Causal Tongue-Tie: LLMs Can Encode Causal Direction, But Their Yes/No Outputs Fail to Express
LLMs encode causal direction internally via probes but revert to commonsense in Yes/No outputs on anti-commonsense items, showing output accuracy alone does not measure causal understanding.
-
LLMs Should Not Yet Be Credited with Decision Explanation
LLMs support decision prediction and rationale generation but lack evidence for genuine decision explanation, requiring stricter standards to avoid over-crediting.
- When Behavioral Safety Evaluation Fails: A Representation-Level Perspective