model activations

3 papers

• Detecting High-Stakes Interactions with Activation Probes
• On Reasoning Strength Planning in Large Reasoning Models
• Projecting Assumptions: The Duality Between Sparse Autoencoders and Concept Geometry