@AmenZwa Not long ago I urged some physicists to distinguish the different kinds of probability, particularly whether we're talking about future events, or merely our ignorance of past events, and how extremely different those are (complete with wikipedia links). It fell on deaf ears. Nay, bordering on hostile ears. (Or maybe there was just silence and I assumed the worst; I don't recall.) Sigh. Anyway, > they place the core probabilistic nature of AI, front and centre. Yes, that's so very important. Yesterday I mentioned my fondness for simulated annealing and genetic algorithms, which are Good Old Fashioned AI (and taken so much for granted that they're not always associated with any kind of AI). Randomness and probability are more obviously at the heart of those, but they're widely important in LLMs and other modern and ancient forms of algorithms, of course. The most obvious place in LLMs is in the heat setting, which must be non-zero to get any kind of variation on "creativity" from them, but less obviously probability shows up in the foundational use of pseudovectors to emulate very high dimensional vectors using more tractable lower dimensional vectors / matrices / tensors. Which is all pretty cool.