There are legitimately AI researchers who are investigating “how can we make models that can be trained on AI generated data”. These researchers know about model collapse (LLMs recursively trained on generated data will degenerate after a few iterations) for over a year now

… I hope I was joking