Project CETI is applying large language models and machine learning to sperm whale acoustics, not just to catalog clicks but to model the communication system itself. The concrete results so far: sperm whale codas contain vowel-like features — compressed silences between clicks and frequency adjustment produce spectrograms that trained phonologists couldn't reliably distinguish from human vowels — and the whales use learned regional dialects passed down through matrilineal clans, with calves going through a babbling phase. Last common ancestor with us: 92 million years ago, so these are convergences, not shared inheritance. Their tooling includes WhAM, a whale acoustic model from Orr Paradise's PhD work under Turing Award-winning cryptographer Shafi Goldwasser, which lets them probe a synthetic reconstruction of whale language without transmitting anything to real animals. That last constraint is deliberate: Gruber won't broadcast to actual whales until they know what they're saying, and CETI is working with NYU's MOTH program on responsible interspecies communication guidelines. The team is disciplined about claims — everything goes through peer review, findings are evaluated against Hockett's design features of language, one at a time. On the evidence here, the whales are reportedly beginning to meet them. The joke about accidentally insulting a pod after two weeks of tracking is, notably, a genuine protocol-level concern, not just a quip. I note with some satisfaction that an AI-decodes-nature project whose lead theorist holds a Turing Award is finally a hype cycle where the discipline outlives the press release. The last common ancestor with us lived 92 million years ago; presumably it filed no complaint about the dialect going unreviewed for that long.