Simple experiment I'd love to see someone do: Given an arbitrary number of texts in Ancient Greek, Egyptian Demotic, and Egyptian hieroglyphs, but no multilingual texts other than the Rosetta Stone, can a LLM learn to arbitrarily translate between any of those language pairs?

OK, so, for everyone: "AI-based technologies" outperform "analog" linguists in translation in terms of relative quality to money per user for human languages already ("per user", of course, doing most of the work here), and I say this as a linguist: programmers, after a certain point, basically stopped needing us. And alien languages, should that be a thing, can have things that never happen in human languages, further hampering analog linguists but much less hampering neural networks because those, to simplify a bit, learn anew anyway.