Large language models (LLMs) might soon be able to develop their own priorities. In that case, we are doomed. But some ...