Wasn't ChatGPT trained on the entirety of Wikipedia? And probably millions of pieces of scientific literature, and arts, and movies and games and and and...
Perhaps the hyperbole of the entire corpus of human knowledge isn't quite technically right, but it's close enough.