alt Hacker NewsGiven that #1 seems to be based almost entirely on stealing from #2, and never paying reparations, I’d say it’s pretty unsustainable.
It’s like saying robbing banks for a living isn’t sustainable and working at a bank is. That’s not exactly a stretch.
#1 may well put #2 out of a living but that isn't the same as stealing and doesn't (at least in and of itself) make it unsustainable. The fact that models were trained on scraped content isn't a matter of technical necessity but rather the path of least resistance (lowest cost in this case). Synthetic data is increasingly used for reasons of quantity, quality, and various technical considerations.