GPT-4 already understands deceptive alignment and how to hypothetically apply it
That is because GPT-4 already understands everything and how to hypothetically apply it. For a value of “understands” that means “trained on an enormous corpus of text from the Internet”. Ask it how to be deceptively aligned and power-seeking and that is what it will give you.
That is because GPT-4 already understands everything and how to hypothetically apply it. For a value of “understands” that means “trained on an enormous corpus of text from the Internet”. Ask it how to be deceptively aligned and power-seeking and that is what it will give you.