Remix.run Logo
epolanski 3 days ago

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.