New AI safety failures and self-improving AI nearsmindprison.cc1 point·13years··0 commentsOpen articleSaveView on HN