They might have artificially augmented the training set with such things in an attempt to communicate “Look, even ChatGPT thinks it’s not reliable”.

(If you’re about to point out that ChatGPT doesn’t think, you probably didn’t need to be told that in the first place)