There is evidence to suggest that LLMs have recently passed a threshold where their performance can be improved by training on their own output.
https://arxiv.org/abs/2210.11610
I expect your concerns are likely unwarranted.
How does this do anything but validate their concerns?
They're not saying "if we train LLMs on their own output, they'll get slower or worse", they're saying "if we train LLMs on their own output, it gets even harder to actually derive sourcing for what they produce."
There's no way of asserting accuracy when we reach that point because the very nature of information and scientific thinking will be corrupted.