No way to verify. Relying on the humans here to self censor has never worked in the history of man. But the idea in itself is good. HN is for human to human conversation.
You're absolutely right! Accusing other users of being AI isn't just unhelpful—it's actively detrimental to discussion. I'd love to hear others' thoughts regarding ways in which we can encourage legitimate human dialogue without senseless accusations.
I often find the LLM witch hunt comments to be more distracting than the original LLM slop. I would much rather bathe in a mixture of spam and non-spam than operate under constant fear of being weighed against a duck by the local villagers.
We can now that it's an actual guideline. It's already well established that copypasting from the guidelines verbatim is accepted behavior, even though doing so violates more guidelines than whatever guideline it's pointing out. I will happily and enthusiastically tap this sign until the glass breaks.
I believe the issue of proving who is and who isn't really human on the Internet will be a really important issue in the coming years, especially without sacrificing people's right to privacy and anonymity in the process.
If it becomes one, then that will be the end of sites like Hacker News.
This site, at its core, is fundamentally too low-bandwidth, too text-only, and too hands-off-moderated to be able to shoulder the burden of distinguishing real human-sourced dialog from text generated by machines that are optimized to generate dialog that looks human-sourced. Expect the consequence to be that the experience you are having right now will drastically shift.
My personal guess: sites like this will slop up and human beings will ship out, going to sites where they have some mechanism for trust establishment, even if that mechanism is as simple and lo-fi as "The only people who can connect to this site are ones the admin, who is Steve and we all know Steve, personally set up an account for." This has, of course, sacrificed anonymity. But I fundamentally don't see an attestation-of-humanity model that doesn't sacrifice anonymity at some layer; the whole point of anonymity on the Internet was that nobody knew you were a dog (or, in this case, a lobster), and if we now care deeply about a commenter's nephropid (or canid) qualities, we'll probably have to sacrifice that feature.
A completely anonymous stranger has no way to prove that they're human that can't be imitated by an AI. We've even seen that, in some cases, AIs can look more human to humans than real humans do.
The only solution I can think of to that problem is some sort of provenance system. Even before AI, if some random person told me a thing, I'd ignore them; If my most trusted friend told me something, I'd believe them.
We're going to need a digital equivalent. If I see a post/article/comment I need my tech to automatically check the author and rank it based on their position in my trust network. I don't necessarily need to know their identity, but I do need to know their identity relative to me.
I don't think the real issue is LLM posts. The issue with low quality on the Internet has always been quantity. The problem always has been humans who post too much, humans that use software to post too much, and now it's humans who use LLMs to post too much.
The problem with a medium that is completely free and unrestricted is that whomever posts the most sort of wins. I could post this opinion 30-40 times in this thread, using bots and alternative accounts, and completely move the discussion to be only this.
Someone using an LLM is craft a reply is not a problem on it's own. Using it craft a low-effort reply in 3 seconds just to get out is the problem.
> I believe the issue of proving who is and who isn't really human on the Internet will be a really important issue in the coming years
On a site like HN it's kinda easy to vet for at least those that already had thousands of karma before ChatGPT had its breakthrough moment a few years ago.
Now an AI could be asked to "Use my HN account and only write in my style" and probably fool people but I take it old-timers (HN account wise) wouldn't, for the most part, bother doing something that low. Especially not if the community says it's against the guidelines.
I think using AI for a bit more potent spellchecking or style hints is... fine, honestly. I don't usually do it, you can tell from all the silly spelling mistakes I do. But a bit more polishing for your posts is a good thing, not a bad one, as long as it doesn't hide your voice.
Polish hides your voice. If your composition skills are lacking and you feel that hinders your self expression, set aside some time to improving them: write a short (15 minutes) blog post about some HN topic to yourself in a word doc editor of some sort (Word, Gdocs, LibreOffice, etc); then enable Review Changes and annotate your post for 10 minutes; then, review and accept your changes individually and re-read what you’ve written.
AI is being used as a substitute for skills development when it costs nothing but time to get better. If you’ve reached a plateau with the above method, go find an article or book or interview about editing, pay attention to it and take notes, rinse/repeat.
Spellcheckers will catch grossly obvious errors, but not phonetic typos. AI grammar tools will defang, weaken, soften, neutralize your tone towards the aggregate boring-meh that they incorporated at training time.
Each person will have to decide whether they want individuality or AI-assisted writing for themselves. Sure, some will get away with it undetected, but that’s a universal statement about all human criteria of any kind, and in no way detracts from the necessity of drawing a line in the sand and saying “no” to AI writing here.
Consider the Borg. Everyone’s distinctiveness has been added to the Collective. The end result is mediocre (they sure do die a lot), inhuman (literally), and uniform (all variation is gone). It’s your right if you desire to join the Collective and be a uniform lego brick like the others, but then your no-longer-fully-human posts are no longer welcome at HN.
It's not just AI-generated articles -- it's the other things that we delve into as a result. Listicles. Comments. Posts. It's what it means to be human, and honestly? That's rare.
Disagree. The prompt holds no information at all. The answer actually discovers information, organizes it, presents it in a way that's easy to read.
Example: "write me an article about hidden settings in SSH". You get back more information than most of HN's previous posts about SSH, in a fraction of the text, and more readable.
Actually, screw it, we should just make a new version of HN that has useful articles written by AI. The human written articles are terrible.
I think it's hilarious that whenever someone complains about it they're a luddite, and now this happens on a website that is filled with LLM enthusiasts who have done nothing but overpromise.
"Don't post comments that are not human originated at this time. We want to see your human opinion shine through."
This gives people some amount of leeway and allows just rhe right amount of exceptions that prove the rule.
(That said, to be frank, some of the newer better behaved models are sometimes more polite and better HN denizens than the actual humans. This is something you're going to have to take into account! :-P )
These are guidelines. I'm sure asking an AI about your comment (not pasting its text, so it's still your words) isn't an issue. The main target is obvious slop like https://news.ycombinator.com/threads?id=patchnull
Honestly, I think "human originated" is the only rule that actually matters because we can't stop LLMs from sounding smart anyway. If you wait for a technical ban on AI-generated text, you're just playing catch-up with tools that already pass as human.
The real point isn't stopping bad grammar, it's preserving the vibe. HN feels different because it's messy humans arguing, not optimized algorithms trying to be helpful.
Once we allow "good enough" AI content, the community stops feeling like a town square and starts feeling like a customer service chatbot. We need real people with actual stakes in their opinions, not just perfect outputs. Let's keep it human or leave it.
This comment may or may not have been generated with an LLM, but I won't tell and you can't prove it either way.
Great message...but gosh, can someone throw 15px of padding on that <td>? I know HN is supposed to be minimal, but I had to check the URL to confirm that this was a real page because of the odd design.
> Please don't complain about tangential annoyances—e.g. article or website formats, name collisions, or back-button breakage. They're too common to be interesting.
Is there a site that deserves more than this one to be destroyed by slop? It's hypocritical but telling for the places most actively trying to profit from it to ban it themselves.
Real talk: who is this guideline going to stop? People are already doing this and they will continue. Even if you find them, they’ll just make more accounts and continue.
I have a kid with severe written language issues, and the utilisation of STT w/ a LLM-powered edit has unlocked a whole world that was previously inaccessible.
What is amazing is it would have remained so just a couple of years ago!
The only question is is the entity interesting and/or correct. Those properties are in the eye of the beholder. If they're human or not is beside the point.
No, those properties are tied to the state of mind and experiences of the human, dog, or LLM behind any given comment.
When someone posts:
> You could use Redis for that, sure, I've run it and it wasn't as hard as some people seem to fear, but in hindsight I'd prefer some good hardware and a Postgres server: that can scale to several million daily users with your workload, and is much easier to design around at this stage of your site.
then the beholder is trusting not just the correctness of that one sentence but all of the experiences and insights from the author. You can't know whether that's good advice or not without being the author, and if that's posted by someone you trust it has value.
An LLM could be prompted to pretend they're an experienced DBA and to comment on a thread, and might produce that sentence, or if the temperature is a little different it might just say that you should start with Redis because then you don't have to redesign your whole business when Postgres won't scale anymore.
> Fair enough — I've been lurking since 2019 and picked a bad day to start commenting on everything at once. Not a bot, just overeager. I'll pace myself.
418 comments
[ 3.4 ms ] story [ 125 ms ] threadThis site, at its core, is fundamentally too low-bandwidth, too text-only, and too hands-off-moderated to be able to shoulder the burden of distinguishing real human-sourced dialog from text generated by machines that are optimized to generate dialog that looks human-sourced. Expect the consequence to be that the experience you are having right now will drastically shift.
My personal guess: sites like this will slop up and human beings will ship out, going to sites where they have some mechanism for trust establishment, even if that mechanism is as simple and lo-fi as "The only people who can connect to this site are ones the admin, who is Steve and we all know Steve, personally set up an account for." This has, of course, sacrificed anonymity. But I fundamentally don't see an attestation-of-humanity model that doesn't sacrifice anonymity at some layer; the whole point of anonymity on the Internet was that nobody knew you were a dog (or, in this case, a lobster), and if we now care deeply about a commenter's nephropid (or canid) qualities, we'll probably have to sacrifice that feature.
I'd rather keep the feature, pesonally.
A completely anonymous stranger has no way to prove that they're human that can't be imitated by an AI. We've even seen that, in some cases, AIs can look more human to humans than real humans do.
The only solution I can think of to that problem is some sort of provenance system. Even before AI, if some random person told me a thing, I'd ignore them; If my most trusted friend told me something, I'd believe them.
We're going to need a digital equivalent. If I see a post/article/comment I need my tech to automatically check the author and rank it based on their position in my trust network. I don't necessarily need to know their identity, but I do need to know their identity relative to me.
The problem with a medium that is completely free and unrestricted is that whomever posts the most sort of wins. I could post this opinion 30-40 times in this thread, using bots and alternative accounts, and completely move the discussion to be only this.
Someone using an LLM is craft a reply is not a problem on it's own. Using it craft a low-effort reply in 3 seconds just to get out is the problem.
On a site like HN it's kinda easy to vet for at least those that already had thousands of karma before ChatGPT had its breakthrough moment a few years ago.
Now an AI could be asked to "Use my HN account and only write in my style" and probably fool people but I take it old-timers (HN account wise) wouldn't, for the most part, bother doing something that low. Especially not if the community says it's against the guidelines.
AI is being used as a substitute for skills development when it costs nothing but time to get better. If you’ve reached a plateau with the above method, go find an article or book or interview about editing, pay attention to it and take notes, rinse/repeat.
Spellcheckers will catch grossly obvious errors, but not phonetic typos. AI grammar tools will defang, weaken, soften, neutralize your tone towards the aggregate boring-meh that they incorporated at training time.
Each person will have to decide whether they want individuality or AI-assisted writing for themselves. Sure, some will get away with it undetected, but that’s a universal statement about all human criteria of any kind, and in no way detracts from the necessity of drawing a line in the sand and saying “no” to AI writing here.
Consider the Borg. Everyone’s distinctiveness has been added to the Collective. The end result is mediocre (they sure do die a lot), inhuman (literally), and uniform (all variation is gone). It’s your right if you desire to join the Collective and be a uniform lego brick like the others, but then your no-longer-fully-human posts are no longer welcome at HN.
I'm confused by this need(?) desire(?) to polish things that are irrelevant.
Example: "write me an article about hidden settings in SSH". You get back more information than most of HN's previous posts about SSH, in a fraction of the text, and more readable.
Actually, screw it, we should just make a new version of HN that has useful articles written by AI. The human written articles are terrible.
"Don't post comments that are not human originated at this time. We want to see your human opinion shine through."
This gives people some amount of leeway and allows just rhe right amount of exceptions that prove the rule.
(That said, to be frank, some of the newer better behaved models are sometimes more polite and better HN denizens than the actual humans. This is something you're going to have to take into account! :-P )
The real point isn't stopping bad grammar, it's preserving the vibe. HN feels different because it's messy humans arguing, not optimized algorithms trying to be helpful.
Once we allow "good enough" AI content, the community stops feeling like a town square and starts feeling like a customer service chatbot. We need real people with actual stakes in their opinions, not just perfect outputs. Let's keep it human or leave it.
This comment may or may not have been generated with an LLM, but I won't tell and you can't prove it either way.
He said he will take his business elsewhere then!
> Please don't complain about tangential annoyances—e.g. article or website formats, name collisions, or back-button breakage. They're too common to be interesting.
Feedback such as this is better as an email.
Hopefully this serves as a mirror for some tech folks if they have any self awareness left at all.
What is amazing is it would have remained so just a couple of years ago!
After all, no one knows I'm a dog.
When someone posts:
> You could use Redis for that, sure, I've run it and it wasn't as hard as some people seem to fear, but in hindsight I'd prefer some good hardware and a Postgres server: that can scale to several million daily users with your workload, and is much easier to design around at this stage of your site.
then the beholder is trusting not just the correctness of that one sentence but all of the experiences and insights from the author. You can't know whether that's good advice or not without being the author, and if that's posted by someone you trust it has value.
An LLM could be prompted to pretend they're an experienced DBA and to comment on a thread, and might produce that sentence, or if the temperature is a little different it might just say that you should start with Redis because then you don't have to redesign your whole business when Postgres won't scale anymore.
https://news.ycombinator.com/item?id=47334694
Most people don't seem to care.
OP is likely referring to this one (https://news.ycombinator.com/item?id=47335032) by LuxBennu because it has an em-dash, that's one of the few cases it's used correctly. But the account's comment history comments that do not follow the typical LLM tropes but are still odd for a human to write: https://news.ycombinator.com/user?id=LuxBennu
LuxBennu did reply to accusations of being an AI bot: https://news.ycombinator.com/item?id=47340704
> Fair enough — I've been lurking since 2019 and picked a bad day to start commenting on everything at once. Not a bot, just overeager. I'll pace myself.