Sparse Autoencoders Reveal Temporal Difference Learning in Large Language Models

Re-assign community