You're still establishing utility as "chief among equals." I have already responded to this premise. The total value of something is the weighted sum of its individual value axis' and those weights are derived from context. Once again, neither of us is discussing a situation in which survival is dependent upon the use of LLM output. We're describing, at worst, CVE remediations and, at best, driver enhancements. While I don't know what your material condition is I feel it is safe to say that you are not likely to die as a result of amdgpu not being up to date. Because of this the word "objectively" that you're using is doing a lot of work.
Regarding the claim that there is no such thing as a "non-utility-feature"; yes there is. There's a whole term for it; " nonfunctional requirements". It's something not uncommonly brought up in product planning. If you mean "utility" in the broadest sense of "does thing" than this conversation is pointless. That definition provides no distinction between "this accomplishes tasks that I want accomplished" and "this is built in such a way as I would want it built". Surely we can agree those are different things?
You are right to say that I find the use of LLMs distasteful even under ideal circumstances. And I have stopped using Linux for that reason. Hell I stopped using Neovim for that reason, I'm very close to getting rid of my browser. If push comes to shove I may eventually go full hermit and use 9front as the thought of using anything that's been touched by them is rather sickening. I'm not sure how much longer I can stand knowing that most everything I see online has been tainted at some point.