Magic and the halting problem — LessWrong