Well, the point I was getting towards is that if you remove the memory-based requirements from the scenario you set out, then it doesn't sound so far off from w

Well, the point I was getting towards is that if you remove the memory-based requirements from the scenario you set out, then it doesn't sound so far off from what something like GPT-3 can already do today. I am trying to reduce your scenario down to the minimal requirements so we can get a better understanding of whether those requirements are achievable by computers or not. If we just rely on "knowing it when we see it" then we might fool ourselves until the point where it's too late to go back.