Four visions of Transformative AI success — LessWrong