Very cool, good way to think about it. I wouldn’t be surprised if non-AGI LLMs help write the code to augment themselves into AGI.

The unreasonable effectiveness of deep learning was a surprise. We don’t know what the future surprises will be.