some kind of quinn on Nostr: the big providers are concerned about model collapse (real data -> stolen to make AI ...

the big providers are concerned about model collapse (real data -> stolen to make AI -> enclosured to "we own the output of what we stole" -> people post enclosured AI slop in the wild -> next round of models end up stealing their own output) so, probably, at least internal to those companies they are watermarking output and consider AI output poisoned.