Of course, there are mistakes in their assumptions about MDN in
that. Like, we only currently use a PR model for data content, not
for prose. That may change going forward, but is not currently true
and won’t be for a good while yet at best.
That said, that sounds like an interesting concept.