The Prospect of an AI Winter — LessWrong