“OpenAI has successfully trained models that exhibit their own ways of thinking, which are legible but different from their initial human-emulating thought patterns.”