Running PaaT: Probe as a Tool for Proprioceptive Language Agents ๐ Generate risk scores for a modelโs next response
Paused Control Reinforcement Learning ๐ Explore LLM token decisions with featureโdriven visualizations
Running 4 CorrSteer: Correlation-Based Steering of Language Models via Sparse Autoencoders ๐งญ Steer language model output by clicking visual layers