Blog topic

Inside the model: interpretability and alignment

What a trained network has learned and how one finds out: probes, the Othello experiment, features and circuits, and the research on aligning models.