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
689 results for “keyword:"mechanistic interpretability"” in ICLR 2027. 142 withdrawn or desk rejected listed apart. clearsort:relevanceratingnewestmost read
[PDF]Towards Eliciting Deployment Behaviour in Scheming Models via Steering-Vector Transfer▼5.8k2.33
[PDF]Small Objects Inside the Grid: Activating Latent Scale Knowledge in Frozen Detectors via Training-Free Grid Reparameterization▼1361.33
[PDF]Linear Readability Does Not Guarantee Writability: Evidence from Controlled Counterfactual Recomposition▼1.2k1.67
[PDF]From Representation to Behavioral Control: An Empirical Study of Fine-Grained and Unified Emotional Signals in LLMs▼4.3k2.00
[PDF]Broadcast, Reinforce, Repeat: Attention Dynamics Behind Repetition Collapse in Diffusion Language Model Forecasting▲1.7k3.33
[PDF]Split, Not Sparse: Ground-Truth Feature Superposition Across Scientific Domains and Model Scale▼1361.33
[PDF]Where You Steer Shapes the Direction: Operator Conditioned LLM Behavior Control and J-lens Interpretation▲3952.67
[PDF]SPECTRAL SHAPE RESIDUAL: A LIGHTWEIGHT SPECTRAL DIAGNOSTIC FOR ATTENTION WEIGHT MATRICES▼1361.33
[PDF]RefusalTrace: Localizing and Validating the Internal Refusal Pathway of LLMs with Sparse Autoencoder Features▲9.7k2.75
[PDF]A Memory That Only Looks Like One: A Hybrid Vision-Language Model Relays the Picture Rather Than Storing It▼1.2k1.67
[PDF]Targeted Retrieval, Compact Representations: How CoT Reasoning Improves Long-Context Counting▼4.3k2.00
[PDF]When Weak Features Hide: Separating Statistical Detection from Sparse Autoencoder Recovery▲6.1k2.00
[PDF]When Are Spectral Modes Causal? Identification Before Intervention in Learning Dynamics▼4.3k2.00
[PDF]Mind the Spike: Mechanisms and Brittleness of Visual Massive Activations in Large Vision–Language Models▼4.5k2.67
[PDF]Programmable Semantic Interventions: Learning Causal Carriers from Behavioral Programs▲6.1k2.00
[PDF]The Model Already Has a Dictionary: Cumulative SVD Features as a Training-Free Alternative to Sparse Autoencoders▲3952.67
[PDF]Same Task Performance, Different Intervention Readouts: Cross-Run Variation under In-Context Rule Aliasing▲3.0k2.33
[PDF]Hidden APIs in Language Models: Discovering Reusable Causal Interfaces from Forked Futures▼4.3k2.00
[PDF]The complexity of understanding: Formalizing Concept Discovery in Mechanistic Interpretability▼1.2k1.67
[PDF]When Does Correction Become Repair? Mechanistic Auditing of Internal Interventions in Tool-Using LLMs▼1.2k1.67
Withdrawn or desk rejected (142)
[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