Interpretability

Old Habits Die Hard: How Conversational History Geometrically Traps LLMs
Same Answer, Different Representations: Hidden Instability in VLMs
The Hitchhiker's Guide to Actionable Interpretability
Automated Interpretability-Driven Model Auditing and Control: A Research Agenda
Interpretability Can Be Actionable
Quantifying the Effect of Test Set Contamination on Generative Evaluations
Context Matters: Analyzing the Generalizability of Linear Probing and Steering Across Diverse Scenarios
Beyond Linear Steering: Unified Multi-Attribute Control for Language Models
Precise In-Parameter Concept Erasure in Large Language Models
Query Circuits: Explaining How Language Models Answer User Prompts