Actually, years of research certainly can be emulated by AIs, in the sense that they have many scanned research papers and can cobble together a solution to many things, say unsolved math problems, from this. So baseline they can seem pretty great. The article is indeed just reinforcing the point that their innovations are less impressive when it's something that really no one has done before. It's not a trivial point for who are used to AI doing so many impressive things.