zlacker

Big surprise that OpenAI's solution here is to train a more powerful language model and ask it how to do alignment.