NeurIPS 2026 reviews are up: reviews, scores, rebuttals and meta-reviews for 7,877 papers. See the highest rated
acceptodds: the papers of ICLR 2027
729 results for “keyword:"mechanistic"” in ICLR 2027. 152 withdrawn or desk rejected listed apart. clearsort:relevanceratingnewestmost read
[PDF]Out of Sight, Still in Mind? Separating Internal and Answer-Level Effects of Retained Key-Value History▼1.2k1.67
[PDF]The Saussurean Representation Hypothesis: A Causal Account of Representational Structures in LLMs▼4.3k2.00
[PDF]Cross-Architecture Universal SAEs: A Shared Sparse Dictionary for Vision and Diffusion Transformers▼1361.33
[PDF]RouteVerify: Routing Interventions for Step-Level Verification in Mixture-of-Experts Models▲3.0k2.33
[PDF]Fireman against Jody: Causal Counterfactual Mirrors for Auditing Strategic Deception in Language Models▼1.2k1.67
[PDF]“Many Are My Names”: The Anatomy of the Assistant and Its Personas via Sparse Autoencoders▼1.2k1.67
[PDF]Language models use both proactive and retroactive mechanisms to make compositional inferences▼4.5k2.67
[PDF]Where Does the Verbalizable Workspace Band Begin? An Audit of the Jacobian-Lens Instruments Across Open-Weight Families▲14k3.00
[PDF]What Happens to the Goal? Distinct Failures of Policy Alignment in Goal Misgeneralization▲3952.67
[PDF]Decodable but Misrouted: Sparse Features Uncover a Readout Gap in Vision-Language Models for Harmful Meme Detection▼1.2k1.67
[PDF]Where Does the Audio Jailbreak Live? A Controlled Frequency-Depth Audit of Audio-Language Models▼1.2k1.67
[PDF]Neuron Sharing in Multi-Task Reinforcement Learning Policies Follows the Geometry of the Task Parameterization▼1361.33
[PDF]An Identifiability Hierarchy for Mechanistic Interpretability: Interpretability Claims Live at Different Levels of Identifiability▼1361.33
Withdrawn or desk rejected (152)
[PDF]Why Benign Post-Training Induces Safety Gate Collapse: Changes in Harmful Activation Geometry Relative to the Inherited Refusal GateWithdrawn1.67
[PDF]Concept Subspaces Compute Beyond the Logit Lens: A Weights-Only Test for Locating Representations Upstream of ReadoutWithdrawn