Danger, AI Scientist, Danger — LessWrong