Hmm. I think I understand where you are going but given that LLMs appear to need injected randomness to work effectively, determinism isn't really on the table so we kind of stop talking about the LLMs we use when we (strictly speaking) consider the case where they could generate deterministic sequences?
replyNot that this in any way takes away anything from your argument that "it's that language is imprecise". That still holds true, but it is on the input side.
Why would LLMs need injected randomness to work effectively? Their nondeterminism isn't by design, it's largely a byproduct of how (batched) inference actually happens on hardware [0]. In case you are referring to the actual token sampling (be it temperature, Top-K or whatever), i wouldn't really see this as part of the LLM itself (And I've frequently seen benchmark sites run them at t=0 anyways)
reply[0] https://thinkingmachines.ai/blog/defeating-nondeterminism-in...