I can't see how being used as training data has anything to do with this problem. Being able to differentiate between the AI slop and the accurate information is the issue.