> Those subreddits label content wrong all the time.

Intentionally if I might add. Reddit users aren't particularly interested in providing feedback that will inevitably be used to make AI tools more convincing in the future, nobody's really moderating those subs, and that makes them the perfect target for poisoning via shitposting in the comments.