The architecture of neural networks that are applied in computers today is actually based on a paper that was originally written in 1943. We know a lot more about human brain neurons than we did in 1943, so it's entirely possible that there are fundamentally better neural network architectures that we will now invent. The GPT four has far superior crystallized intelligence to any human right because it's a computer and so it the other thing is happening is it just knows about so many more things.