My reading is that the intent is to have a human voice behind the text.
Monitor and see how it goes I guess!
The short version is that we included it to protect users who don't realize how much damage they're doing to their reception here when they think "I'll just run this through ChatGPT to fix my grammar and spelling". I've seen many cases of people getting flamed for this and I don't want more vulnerable users—e.g. people worried about their English—to get punished for trying to improve their contributions. Certainly that would apply to disabled users as well, though for different reasons.
Here are some past cases of these interactions: https://hn.algolia.com/?dateRange=all&page=0&prefix=false&qu....
Edit: uni_baconcat makes the point beautifully: https://news.ycombinator.com/item?id=47346032.
Most rules in https://news.ycombinator.com/newsguidelines.html have a lot of grey area, and how we apply them always involves judgment calls. The ones we explicitly list there are mostly so we have a basis for explaining to people the intended use of the site. HN has always been a spirit-of-the-law place, and—contrary to the "technically correct is the best correct" mentality that many of us share—we consciously resist the temptation to make them precise.
In other words yes, that bit needs to be applied cautiously and with care, and in this way it's similar to the other rules. Trying to get that caution and care right is something we work at every day.
I would wager that this use case is much more prevalent than ones where the LLM changed the comment significantly enough to change one's voice.
I never copy/paste from an LLM into HN. Everything is typed by myself (and I never "manually" copy LLM content). I don't have any automatic tools for inserting LLM content here.[1]
Always, always, always keep in mind that you don't notice these positive use cases, because they are not noticeable by design. So the problematic "clearly LLM" comments you see may well be a small minority of LLM-assisted comments. Don't punish the (majority) "good" folks to limit the few "bad" ones.
Lastly, I often wish we had a rule for not calling out others' comments as "AI slop" or the like.[2] It just leads to pointless debates on whether an LLM was used and distracts far more than the comment under question. I'm sure plenty of 100% human written comments have been labeled as LLM generated.
[1] The dictation one is a slight exception, and I use it only occasionally when health issues arise.
[2] Probably OK for submissions, but not comments.
I was thinking of calling this service "Dang It."
You say you want hear posts in other people's voices but I'm pretty sure that if I did this that the people who used it would find greater acceptance of their comments than if they just posted them as they originally wrote them.
One dynamic I don't think has yet been given its due: while AI is training on us, we're also all getting trained on it—that is, the hivemind's pattern-matching ability is also growing. We're heading up the escalation ladder in a paattern-matching race.
But that name is hilarious!
It's a bit different when specific cases come up because then there's a chance to talk about it, add clarifying comments, etc.
For me that link says:
> Error: Forbidden
> Your client does not have permission to get URL / from this server.
https://hn.algolia.com/?dateRange=all&page=0&prefix=false&qu...
How the hell does does this place exist right now with all that is going on. I dont know much about YC, but they don't seem that humane..
Also writing a draft in Google Docs and accepting most [2] of the corrections is fine. The browser fix the orthography, but I 30% of the time forget to add the s to the verbs. For preposition, I roll a D20 and hope the best.
I'm not sure if these are expert systems, LLM, or pingeonware.
But I don't like when someone use a a LLM to rewrite the draft to make it more professional. It kills the personality of the author and may hallucinate details. It's also difficult to know how much of the post is written was the author and how much autocompleted by the AI:
[1] Remember to check that the technical terms are correctly translated. It used to be bad, but it's quite good now.
[2] most, not all. Sometimes the corrections are wrong.
This makes me think of something: are nonnative English speakers tempted to use LLMs to correct grammar because mistakes like this actually make the writing unintelligible in their native language? For example, if I swap out the "For" in this sentence for any (?) other preposition, it's still comprehensible. (At|Of|In|By|To|On|With) example, ...
All of them are comprehensible, but are wrong, nobody would use them. If a foreigner use them (the translated version) people will understand them, but it will sound odd. Depending on the context, people will correct it or just go on.
Perhaps "As" or "Like" are better, still not 100% accurate but almost.
"Hoy no comí nada." -> literally "Today [I] did not eat nothing." but should be "Today, I didn't eat anything."
May be confusing, but people may ask for clarifications.
---
Another problem is that some words have different meaning in each country, like
"I will pick an apple." -> to es-es "Voy a coger una manzana" but in es-ar it means "I will fuck an apple."
May be confusing too, but if you have a strong Spanish accent or American-that-Learned-Spanish-from-Spain accent, people will chuckle and go on.
Strong disagree on author voice. Vomit blows.
I think better to let recipient use full-text translation if that is necessary.
Anyway, the autotranlation saved a lot of time for the most common words and switching the noun-adjetive order.
But like dang said ... I do not have time to fight this battle when I have only 10 minutes :)