what can language model embeddings tell us about whales speech, and decoding ancient texts?      (on The Platonic Representation Hypothesis and the idea of *universality* in AI models)