An essay distills lessons from AI safety failures into alignment guidance. It argues that safety depends on more than training objectives, urging developers to examine real-world exploits and rethink where alignment research goes next.
Opening Kapyn…