Large Language Models will be Great for Censorship — LessWrong