Problem is that so much current content is already AI generated that its becoming difficult to train LLM models using the Internet. Going forward, either we find a way to train AIs on AI content without driving it insane, or we keep dipping back into the only deep well of guaranteed human content way into the future.