Transformative AI and Compute [Summary] — LessWrong