If it was genuinely useful, we would've long reached the point where you train a model on a previous one's output in an ever improving loop. But this doesn't a

If it was genuinely useful, we would've long reached the point where you train a model on a previous one's output in an ever improving loop. But this doesn't actually work.