> huge volumes of plausible blithering that's factually wrong

That describes large parts of the internet, and we trained these models on that, so what did we really expect? ;-)