I'm glad the author mentioned the AI alignment issue that could arise from this. AI systems are fundamentally unexplainable, such as the Loab cryptid that occurs in DALL-E-2. I wonder what sorts of similar perverse instantiations we will see with these types of models.