Somayeh Sojoudi

Reinforcement Learning via Value Gradient Flow

Revising and Falsifying Sparse Autoencoder Feature Explanations

Generalization or Hallucination? Understanding Out-of-Context Reasoning in Transformers