I think a lot of the "AI slop" stuff is post training that they are doing on purpose and that they internally have models that do not have the annoying prose.