It does seem like limiting new training material to only good quality information would be prudent rather than slurping up entire GPT-produced spam sites. Sure,

It does seem like limiting new training material to only good quality information would be prudent rather than slurping up entire GPT-produced spam sites. Sure, LLMs will have been used to refine and assist much of it from here on out, but I'd like to think theses, newspapers, etc won't just be produced entirely by unsupervised robots and not even fact-checked. If that's the case we'll have bigger problems.