At this point, I think google, openai, anthropic, etc already realise this and are just trying to pretend this isn't true. I even think some C-suite who are not in AI companies but are boosters know this too. This has been true since 2022 but they're hoping (likely correctly) that governments won't move fast enough to protect the IP of the actual productive class.

I think the long term reality is that the models still need training data so they fundamentally do need new writing/code/art to train on, and even then the usual issues like hallucination will still be with us. It's just the moment that actually hurts the (already questionable) profitability of the model peddlers, they will have gotten their IPOs and they can safely jump ship and the ultimate mess can be passed to the softbanks, the temaseks, and the governments of the world to clean up for them. What the future holds after the crash I'm not sure as the models won't disappear (especially now that the stolen data is already crystalised in open source models) but in the near term the mass theft that constitutes llms will become more and more understood even amongst the PMC and that in order to remain viable, you need the productive to keep producing, and unlike LLMs, you can't force them to do it without payment.