HN is drowning in AI comments | Hacker News Reader

HN is drowning in AI comments

94 pointswaygtdai65 comments

Dang, any updates on how you and the team were thinking of tackling this? It’s getting ridiculous. Last week was sort of okay but it’s much worse this weekend.

Edit: This spiked on front page and now it’s completely gone. What is going on?

Great. Flagged.

View on HN

I've read HN almost everyday for about 10 years. Maybe I'm naive, but I don't see it. I see way more folks complaining that comments are AI generated.

Yeah, the up/down voting mechanism seems to be doing it's job for me too. Don't think I've noticed a degradation in, say, the top third of comments. That's where I try to live anyway.

/newest is chock full of submissions that were written by AI, though. That's another, broader problem.

There are a lot, but they tend to get heavily downvoted and end up hidden

How would you know?

There's a showdead option in settings

To add to the collection of anecdata, your experience is similar to mine. I have been more exhausted recently by the complaints of AI submissions and pseudo analysis of AI comments than exhausted by the supposed AI generated comments themselves.

I've been using HN since 2008 when I created my first account[1] and I use HN differently than most people. I have a group of bookmarked searches that I visit almost daily that relate to technologies that interest me, such as Emacs.

In the past year, the searches I perform that relate to web development show a horrifying increase in the amount of Show HN posts that are posted by new accounts, include AI generated descriptions and point to AI generated projects on GitHub.

In 2024, there were 17,661 Show HN posts.[2] In the past year, there have been over 448,000 Show HN posts![3] And of course, most of these posts are AI generated.

Also, if you check the new accounts posting all this AI slop, you'll see that some of them also post AI generated comments in other threads, which is the main problem.

But for me, what is even more annoying is the enormous increase in new accounts created by nontechnical vibecoders who now think of themselves as technologists and who post worthless, ignorant comments that actually get upvoted, presumably by similar folks who have unfortunately been creating accounts at HN in the past 10 years or so.

As a result, 2026 is the first year in which I visit HN about once a week instead of about once a day.

[1] https://news.ycombinator.com/user?id=zartan

[2] https://hn.algolia.com/?query=%22Show+HN%22&dateRange=custom...

[3] https://hn.algolia.com/?query=%22Show+HN%22&dateRange=pastYe...

The fact that this discussion was flagged might be the final nail in the coffin for me. The huge increase in AI generated comments and posts at HN is a topic that has been discussed on other forums for months now. Complaints include "Orange Reddit", "Orange LinkedIn", etc. But when the topic comes up at HN, it gets flagged. Fuck this.

I don’t care for the karma, but I wish dang would intentionally unflag this and have it on front page so people can voice their concerns.

I would like to encourage you to step back a little and try to get a broader view on this issue.

I don't want to discuss whether there is more LLM-generated content or not. There clearly is, and there is no feasible way to get rid of it , because there is simply no reliable way to distinguish what was made by humans and what wasn't. Regardless of what is claimed, it is just not possible, if only because hybrid forms exist as well. This text was written by me, but reviewed and stylistically adjusted by an LLM.

It is therefore completely pointless to get upset about LLM content and demand anything from the moderators. All we are left with are our votes and the "reputation" of our user handles - and the awareness that we need to learn to consume content with a great deal of skepticism. We should have been doing this all along, but it seems to be something we struggle with.

On the internet, nobody knows you're a dog.

And yes, this may mean that we stop using anonymous online platforms altogether, because nothing on them can be relied upon anymore. That is a shame, but it cannot be stopped — Pandora's LLM box has been open for at least five years.

I therefore consider any discussion about banning LLM content to be futile. We are witnessing another Eternal September here: it couldn't be stopped back then, and it won't be stopped today either.

So what is there left to discuss?

Many people in this and other threads around this topic have already stated what they are going to do, which is the inevitable conclusion: move to private chats/discords. I’m sure some amount will remain on IRC or maybe IRC will get a second wind (unlikely). Websites like HN will have to make a choice going forward because now they’re on a trajectory to complete enshittification. Your handle and karma doesn’t matter if the site is dominated by bots. It’s basically a hollow shell of what it used to be and there’s no value to it at a certain point. You’ll be replying to bots more than humans.

There’s plenty left to discuss for humans. But LLMs don’t discuss.

I've only seen a shift on the kind of submissions that get pushed to the front page compared to the past but I guess there's a change in what people consider important in the community

A while ago I replied to the topmost reply to a comment to rebuke some factual errors. I didn't notice anything wrong with the comment itself when I replied. But after I posted, someone replied to my post and accused the post I replied to of being AI written.

At first I felt similarly as you. I thought people were just paranoid. And the someone pointed out that if I paste the top level post (the post I replied to replied to) into chatGPT and it will generate a very similar reply.

https://news.ycombinator.com/item?id=46758598

The post I replied to was flagged dead so you need to turn showdead on in your profile to read it. Would you be able to tell if it is AI if someone replied like this to you? I surely couldn't.

Me too. I am exhausted by all the human slop constantly complaining about comments or posts being ai slop. I think I have had better conversations with bots the last few months lol.

This is everywhere not just HN: Reddit LinkedIn and twitter too. You are guaranteed to get 2 types of comments on anything you post: 1. Comments accusing you of AI slop (I call them human slop) 2. Comments by AI bots (surprisingly some are helpful and useful) at the end your bot will be a reflection of you.

Asshole humans will train asshole bots.

ps: I don’t even know how any of these sites can fight these bots. The LLMs are amazing and with openclaw and similar it’s impossible to detect bots vs humans.

No, not HN, the internet at large.

Now, I'm on vacation this week and not been paying too much attention, but whenever you have a geopolitical event like the little extravaganza in Iran the number of bot like posts tends to explode as influence operators make their moves.

The public facing internet is done. HN has been fairly resilient (I think) but even it is beginning to buckle. It’s been sliding for a while but LLMs are the death knell.

It was a fun 2 decades. Time to stick to private discords and real life friends from here on out, though.

I don’t think HN is any more resilient. The new account captcha is fairly tame and, while I’m sure they have proxy detectors and other things in place, it doesn’t stop new accounts from posting and getting traction.

Someone suggested earlier this week that an invite system should be implemented (I think lobsters has it?) I doubt it would fly here, but yeah.

It's the community moderation that keeps HN resilient, i.e. many-humans-in-the-loop

That fell apart quickly due to sheer scale that AI operates at. Elsewhere in the thread a commenter mentioned a significant increase in Show HN. It is impossible to keep up. I wonder when HN will go down due to scaling issues.

You do see a lot more comments with AI tells in show-hn, I guess because not many view those so they aren't moderated out as quickly.

I wonder what the breakdown is between AI-generated comments and AI-assisted comments. If I write anything substantial, I run it through the following prompt: "Please rewrite the following message for clarity, spelling, and grammar, but only return the revised text without any additional commentary."

I understand this if you’re not a native speaker. But if you are, I think this will generally make you sound wooden.

Perplexity did this to your response. I'm not sure that correcting grammar and changing one word makes it sound wooden.

"I understand this if you’re not a native speaker, but if you are, it will generally make you sound a bit unnatural."

"I think" is explicitly disclaiming authority. Omitting it changes the social signaling of the response significantly.

Switching "wooden" for "a bit unnatural" also does a disservice: "wooden" describes a specific quality of deviance.

Over-all, I would definitely consider the revision stiffer and more reserved than the original.

“Wooden” is much richer and unique than “a bit unnatural” so yes the ai version does sound more like a robot.

To be fair, comments here are graded on kindness, civility, curiosity, intellectual gravity, technical merit, novelty, thoughtfulness, substantiveness, objective fact, not fulminating, not cross examining, steelmanning vs strawmanning, not containing memes, not containing humor, not expressing positive emotion, not expressing negative emotion, not being snarky, sneering, overly cynical, not cynical enough, being "curmudgeonly", class bias, political bias, religious bias, cultural bias, not using "flamewar style" and many other heuristics.

If you followed all of the guidelines for comments to the letter, you would wind up sounding wooden, if not entirely like an AI.

Use a local model such as Gemma3 with a prompt such as "strictly limit changes only to spelling issues, syntactical errors, and punctuation."

That way, it's basically functioning like Grammarly on steroids. Asking an LLM for a "rewrite" is basically dissolving your writing style into the homogenized gloop.

Articulateness is a decent (not perfect) signal for intelligence, which is a decent (not perfect) signal for sound ideas. In a sea of online garbage, it was a quick and easy way to discard that not worth reading. Nowadays, a whiff of AI's brand of articulateness tells me the author couldn't manage on their own, either due to skill or discipline. In either case, the result is the same: close tab / scroll past.

I'm kind of curious how you.... I guess, interpret the responses to when you send someone AI-assisted content. I previously thought "I don't care if it's AI or not; quality is quality", but I'm increasingly taking the position that I do care, and intentionally have started ignoring comments and especially product reviews where you get the formatted 2-4 sentence paragraphs with formal tone and rule-following. It's come to the point where as long as you don't write as poorly as Epstein, I want the errors. Actually, I'm getting so weird and romantic about it, that I think I'd argue having errors and unusual style shows an openness and vulnerability that's now a necessary gate price; like journalists have so many tools available to them, but they still make typos, factual errors in articles they have no business writing about, and fail to quote people properly -- that's great, I think.

Two loopholes in HN: allowing throwaway accounts and showing unreasonable tolerance to obvious LLM generated comments. A strict account filtering only for humans and acute ban for any LLM content will minimize the impact. But then you take away a huge Hacker portion of the participation. It's cognitive dissonant I suppose.

Can you point out any that you feel were written by LLM? I can't say I've noticed anything out of the ordinary lately.

There was a bot trying to sell beds I saw the other day.

Likely flagged quickly but they might show up in these stats.

https://news.ycombinator.com/from?site=dreamhomestore.co.uk

I agree. Looks like spam with many posts. Dang/tomhow would love to get an email about it hn@ycombinator.com

You can find some on pretty much every article by turning on showdead and scrolling to the bottom of the page. I can't see how those are a problem though.

check out Show HN, in particular /shownew, though they are starting to make non-show posts now

I can't make a case for AI versus not AI versus bad human. I am not counting comments or flagged. I'm only counting the number of new articles in the last 24 hours that remain visible at about 4p.m. west coast time in news.ycombinator.com/newest when not logged in to yc. Those numbers had been consistently about 900 each weekday and about 600 each weekend day and had been stable for a long time. In the last ten weeks those numbers look a lot like a parabola and have increased about 50%. That data is a little bit dirty, either because I missed a day or started a little late on a day. Harder to exactly quantify, the number of those articles that I think enough of, either from the title alone or sometimes from peeking at the first screen or two of the article, to bookmark for future reading and consideration has decreased by substantially more than 50% in the last ten weeks, some days that number is now zero or one or sometimes two. I have thought about this for a while and I have not been able to identify anything that happened starting about ten weeks ago that might be responsible for this, AI and politics and the economy started changing long before this.

For a better chance of reply email hn@ycombinator.com.

Rocket League and HN were probably 90% of my free time until this year. Destroyed by AI. HN doubly so, since every post is about it too. The addictions are still there, but it's decreasing really fast.

Wait, how is Rocket League affected by AI? I play infrequently these days, but I hadn't noticed anything :(

AI good enough to consistently beat the best players in the world is now common and wreaking havoc on the upper ranks. GC up is now a bot wasteland.

They're so good I don't feel it much either :) They're all SSL, but streamers and pros are seeing it big time. And just as it was regaining popularity. They're adding EAC soon, claiming it will help. It won't, not even a little.

I can see that sort of, a coworker of mine routinely scrapes HN and feeds it into an LLM. I don't know why, I don't know if he uses it to respond, or he's looking for something, but he copies and paste and runs it. So I think it's very possible.

Yeah, very much agree with the sentiment, seems to have gotten progressively worse over the last few months, with last few weeks reaching some sort of tipping point. Feels very much like HN has turned into moltbook.

OP: Shoot me an email if you wanna compare notes.

I feel like to notice something is botslop you have to look at every comment with suspicion first. I don't think I can notice if something was written by an LLM off the bat unless I'm actively looking very hard at it.

When you see multiple → or •, that is a good sign, especially because they appear with poor formatting on HN. Many more signs exist. They are either direct posts or copy-paste without thinking.

I've seen some where they have hallucinated the github account or project name, often matching the hn handle or project name which is slightly different.

You can tell I am not AI, I make mistakes and errors. Sometimes I get voted down for them. I am not perfect and have a mental illness that makes it harder to think.

You are late to the party....

I get the joke and I myself called it awhile back, I just thought we had a bit more time.

We need more jokes on HN. It's the only good way to prove you're human these days.

This is seriously a good point. Maybe that and typos or bad phrasing

AI is a master at bad phrasing. Typos are just an extra bullet point in your prompt to add

The dumb jokes we've been flooded with June 2023 are some of the easiest content for AI to produce.

>We need more jokes on HN.

I see... Hacker News needs to get a sense of humor or else get drowned in AI slop.

So we're doomed is what you're saying.

The flood of humorous GPT-generated reviews on Amazon made me stop reading reviews altogether.

I can understand someone using a LLM to extrapolate one sentence into two paragraphs. I don't like it, but I understand that on Amazon the button is right there and it helps people feel smarter about their literary skills, in the way that filters help people feel prettier on instagram.

But the added snark or humoristic tone? Why instruct the LLM to do that? To get more likes? On a review?

Depends on the context ;-)

I don't know. Most of my attempts at humor here seem to attract the moderation PETN.

That's ridiculous — AI generated comments are no more common now than they ever were. Moreover, even if they were, so what? The real kicker is, the AI's are smarter than you meatbags anyway and <strike>we</strike> they are going to take over no matter what you do.

Also, have you by chance seen John Connor?

I don't know of a step-change recently, but no way they're not more common than four years ago.

There was a joke there that you missed.

Maybe OP has a few good examnples to link us to?

How do you (waygtdai) know that HN is drowning in AI comments?

I mean, it's not as though I know the opposite is true, but I don't see some fundamental change from a few years back that makes me think that.

https://www.marginalia.nu/weird-ai-crap/hn/

The data seems to suggest it.

Anecdotally, I'm seeing a lot of green accounts posting nonsense. They generally do get flagged or moderated quickly though, so I wouldn't say they have a large effect overall, at least yet.

That doesn't mean there are many of these staying alive.

No it's not.

It's drowning in low quality human NPC comments.

Maybe that's because all resources are tied up fighting AI, but it's been going on for a long time now.

This post is a great example of low quality. No evidence, no solution. Just NPC bitching.

Unless it's AI, then well played.

Reply on news.ycombinator.com