> is what style guides that go into the minutiae of comma placement try do do

Eh. There might be a tacit presumption here that correctness isn't real, or that style cannot be better or worse. I would reject this notion. After all, what if something is uniquely crap?

The basic, most general purpose of writing is to communicate. Various kinds of writing have varying particular purposes. The style must be appropriate to the end in question so that it can serve the purpose of the text with respect to the particular audience.

Now, we may have disagreements about what constitutes good style for a particular purpose and for a particular audience. This will be a source of variation. And naturally, there can be stylistic differences between two pieces of writing that do not impact the clarity and success with which a piece of writing does its job.

People will have varying tastes when it comes to style, and part of that will be determined by what they're used to, what they expect, a desire for novelty, a desire for clarity and adequacy, affirmation of their own intuitions, and so on. We shouldn't obfuscate and sweep the causes of varying tastes under the rug of obfuscation, however.

In the case of AI-generated text, the uncanny, je ne said quoi character that makes it irritating to read seems to be that it has the quality of something produced by a zombie. The grammatical structure is obviously there, but at a pragmatic level, it lacks a certain cohesion, procession, and relevance that reads like something someone on amphetamines or The View might say. It's all surface.

dsign's callout of the minutiae of comma placement is a useful starting point because it's largely rhythmic, and monotony, you could say, is the enemy of rhythm. My go-to example here would probably be the comma splice, which is inflicted on people learning to write in English (while at the same time being ignored by more sophisticated writers) but doesn't exist in e.g. French.

I can be convinced that different spaces need different styles. But, correctness intrinsically emanating from language? That one is not an absolute, unless one happens to be a mathematician or GHC the Haskell compiler or any of the other logical automatons we have and that are so useful.

Language and music (which is a type of language) are a core of shared convention wrapped in a fuzzy liminal bark, outside of which, there is nonsense. An artist, be it a writer or a musician, is essentially somebody whose path stitches the core and the bark in their own unique way, and because those regions are established by common human consensus, the artist, by the act of using that consensus, is interacting with its group. And so is the person who enjoys the art. So, our shared conventions and what we dare call correctness are a medium for person-to-person communication, the same way that air is a medium to conduct sound or a piece of paper is a medium for a painting.

Furthermore, the core of correctness is fluid; language changes and although, at any time and place there is a central understanding of what is good style, the easy rules, such as they exist, are limited and arbitrary. For example, two different manuals of style will mandate different placements of commas. And somebody will cite a neurolinguistics study to dictate on the ordering of clauses within a sentence. For anything more complex, you need a properly trained neural network to do the grasping; be it a human editor or an LLM.

> The grammatical structure is obviously there, but at a pragmatic level, it lacks a certain cohesion, procession, and relevance that reads like something someone on amphetamines or The View might say. It's all surface.

Somebody in amphetamines is still intrinsically human, and here too we have some disagreement. I can not concede that AI’s output is always of the quality produced by a zombie, at least no more than the output of certain human editors, and at least not by looking at the language alone; otherwise it would be impossible for the AI to fool people. In fact, AI’s output is better (“more correct”) than what most people would produce if you forced them to write with a gun pointed to their head, or even with a large tax deduction.

What makes LLMs irritating is the suspicion that one is letting one’s brain engage with output from a stochastic parrot in contexts where one expects communication from a fellow human being. It’s the knowledge that, at the other end, somebody may decide to take your attention and your money dishonestly. That’s why I have no trouble paying for a ChatGPT plan—-it’s honest, I know what I get—-but hesitate to hire a human editor. Now, if I could sit at a caffe with said editor and go over their notes, then I would rather do just that.

In other words, what makes AI pernicious is not a matter of style or correctness, but that it poisons the communication medium—-it seeds doubt and distrust. That’s why people—-yours truly—-are burning manuals of style and setting shop in the bark of the communication medium, knowing that’s a place less frequented by LLMs and that there is a helpful camp filled with authoritative figures whose job of asserting absolute correctness may, perhaps, keep the LLMs in that core for a little longer.

Those are workarounds, however. It's too early to know for sure, but I think our society will need to rewrite its rules to adjust to AI. Anything from seclusion and attestation rituals for writers to a full blown Butlerian Jihad. https://w.ouzu.im/