I’ve read this article once already. It’s no curse; it’s a natural consequence of the probabilistic nature of how those models work. Avoiding it would require an infinitely large pool of human produced content (not going to work) or to at least partially ditch the statistical approach.