This is a pretty common conflation, honestly.

I think people assume that, because computers struggled with it once, there's some deeper difficulty to language. There isn't. We've known since the 1950s that language has a pretty low entropy. So it shouldn't surprise people that text prediction is actually really really good and that the real barriers are ingesting and efficiently traversing.

ETA: arguing with people about this on Reddit does make me want to bring back my NPC Theory of AI. After all, it's possible that a Markov chain really does have a human-level understanding because the horrifying truth is that the people around you are mostly just text prediction algorithms with no real internal reality, too.