Don't we also eventually run into a situation where so much of the online content is AI generated that LLMs are being trained mostly on their own data. In that

Don't we also eventually run into a situation where so much of the online content is AI generated that LLMs are being trained mostly on their own data. In that situation, it seems we could get stuck in a self-reinforcing loop rather than branching out with creativity and ingenuity, because even if such content still exists out there it risks being drowned out.