The Ultimate Newcomb's Problem — LessWrong