The case for a negative alignment tax — LessWrong