My understanding of transformer technology (as present at least) is that it's very hard to update. So a (near) real time GPT3.5/4 isn't possible without a fundamental technological breakthrough, which based on various interviews I'm fairly sure hasn't happened yet.