I think this all convinces me that neither of the things we're discussing here is "neat". Certainly contemporary LLMs don't seem to fit that definition at all, despite being "mathematical".
It's not really up to you to redefine meaning of well known long standing historic technical terms. They're well defined, widely understood, frequently discussed terms that you would learn if you studied the history of AI, read any of the numerous papers and books and discussions about it that I already cited and quoted, and learned about the works of CogSci and AI pioneers like Robert Abelson, Roger Schank, Marvin Minsky, and others who've written numerous papers about it.
The wikipedia page about Neats and Scruffies that I linked you to is in my opinion well written, clearly defines the meanings each term, and presents plenty of evidence and citations and background. I'll give you the benefit of doubt that of course you've already read and understand it, so if you disagree with the history and citations on the wikipedia page and all the original papers and books and people cited and quoted, and can present better evidence and arguments to prove that you're right and they're all wrong, then you are free to go try to rewrite history by sharing your own definitions and citations, and correcting the errors on wikipedia. Good luck! I suggest you start by writing suggestions and presenting your evidence on the talk page first, instead of just directly editing the wikipedia page itself, to see what other experts in the field think and achieve consensus, or else it will likely be considered vandalism and be reverted.
You seem to be missing the point that the world is not strictly black and white, and ever since the terms were originally coined, the people who defined them and many other people have strongly recommended fusing both the "neat" and "scruffy" approaches, and LLMs actually do incorporate some ad-hoc "scruffy" aspects into their mathematical "neat" approach, and that's why they work so much better than simple perceptrons or neural nets. But they are still much more "neat" than "scruffy", and combining the two approaches does not flip the meaning of the two terms. I just discussed the fusion of scruffy and neat here, and quoted the original 41-year-old essay from 1982 by Robert Abelson that defined the terms and recommended fusing the two different approaches:
But before you go off and edit the Neats and Scruffies wikipedia page with your own definitions, please take the time to read the original essay by Robert Abelson that defines the terms first, like I did. In the link above, I cited it, tracked down the pdf, and quoted the relevant part of it for you, but you should probably do your homework first and read the whole thing before editing the wikipedia page about it. But be aware that it uses a lot of other technical terms and jargon that have well known definitions to practitioners in the field, so the common layman definitions of words you learned in grammar school may not apply.
Cyc is clearly the paradigm of "scruffy" and like biology, and perceptrons and neural nets are clearly the paradigm of "neat" and like physics, and that's how those terms have been widely used for more than four decades.
There's no need to be rude! I haven't proposed alternate definitions, I have merely commented on how successful the words used for the AI jargon terms fit with the words in the english language. (And I've appreciated your links on the historical development, which would not have sought out on my own.)
I think what's interesting in the jargon vs. plain definition tension here is related to what you noted in this most recent comment. It's that the words "neat" and "scruffy" - that is, just the english words, not the AI jargon terms - are not really symmetrical. A scruffy thing can easily become more neat while remaining scruffy, but introducing scruffiness into a neat thing tends to just make it scruffy. Neat is more totalizing.
So you say LLMs still fall into the "neat" camp - AI jargon this time - because of their mathematical core and lineage, and that's fair enough. But you also say that they incorporate "scruffy" techniques - jargon again - and I think that makes them - switching to the english words here - seem pretty scruffy, because the scruffy techniques are themselves scruffy, and incorporating all these different techniques is itself a scruffy thing to do.
