Enhancing Automated Interpretability with Output-Centric Feature Descriptions Paper • 2501.08319 • Published Jan 14 • 10