But if the post was generated through a long process of back-and-forth with the model, where significant modifications/additions were made by a human? I don't think there's anything wrong with that.
I do agree with your core point - the thinking is what matters. Where I've found LLMs most useful in my own writing is as a thinking tool, not a writing tool.
Using them to challenge my assumptions, point out gaps in my argument, or steelman the opposing view. The final prose is mine, but the thinking got sharper through the process.
Especially since everyone harps on about it but never provides concrete evidence. If your thinking has sharpened, surely you can find a way to demonstrate how.
I suspect it's one of those things where the user thinks they have improved but the reality is different.
But AI-generated content is here to stay, and it's only going to get harder to distinguish the two over time. At some point we probably just have to judge text on its own merits regardless of how it was produced.
I do note that recently, I wonder what was the point the author wanted to make more often only to then note that there are a lot of what seems to be the agreed on standard telltale signs of excessive AI usage.
Effectively there was a lot of spam before already hence in general I don't mind so much. It is interesting to see, though, that the “new spam” often gets some traction and interesting comments on HN which used to not be the case.
It also means that behind the spam layer there is possibly some interesting info the writer wanted to share and for that purpose, I imagine I'd prefer to read the unpolished/prompt input variant over the outcome. So far, I haven't seen any posts where both versions were shared to test whether this would indeed be any better, though.
I do think there's a great deal wrong with that, and I won't read it at all.
Human can speak unto human unless there's language barrier. I am not interested in anyone's mechanically-recovered verbiage, no matter how much they massaged it.
What many people don't seem to realise, though, is that there's a vast yawning difference between "take this text and translate it from language A to language B" and "take this text and summarise it" -- which bots that can't even count simply cannot do. If a model can't count, it can't work out which subjects come up more or less frequently, for instance, and therefore which are more important.
If it were two humans speaking through a communicator in real time, you can work out what the bot is saying and get around its foibles.
I have personal experience of this. Way back in about 2008 or so, there was a bot known as the Salmon:
https://web.archive.org/web/20081220145156/https://en.wikipe...
It called itself (adjective-beginning-with-S)(Salmon), such as @SuspiciousSalmon.
It scanned Livejournal and looked for profiles with AOL AIM addresses. If 2 accounts posted LJ updates more or less simultaneously, it knew you were both online and messaged you both, typically with an extremely sexual message, and then left you to talk. The snag was, the bot gave you both the same fake name, and any attempt to give your real name or LJ or AOL screenname was replaced with the bot alias.
It got me and a random woman in Baltimore and it took us concerted effort to find out who each other really were. The fun bit is that we're still in touch and occasionally chat or exchange emails.
With human thought, in real time you can bypass limitations introduced by a bot. Do it to static text, including program code, and you can't.