ChatGPT has been around for 4 years at this point. Not very long, but I’ve heard of the ‘imagine what it’ll do in one year’ spiel quite a few times by now.
Transformers were treated like any other ML technique until Sutskever decided to just go big on training it. That it can look like a compelling simulacrum, I am not arguing, but this thing left the ivory tower of research prematurely and recklessly. We are all going to pay for it.