I think a more likely theory is that those accounts are not actual automated bots, it's just that HN used to mostly be for, well, hackers, but now more and more of the audience are AI startup bros who are also more likely to use AI to "phrase their thoughts" because they don't see anything wrong with that.
This is probably the most likely theory. I can't stand reading HN anymore. It is so filled with AI Koolaid and startup tech bros that I find it unbearable. I mean, it has always had a high level of that sort, but it has both gotten much worse and my tolerance for it has evaporated.
I haven’t used it for a long time but I had it an an RSS feed.
Every time I open it I feel like it’s just as astroturfed as some people claim reddit is, I stopped going on there after a while.
That matches what I've seen, one recent post that was at the top of the front page used clearly AI generated text to promote a (vibecoded) project. The OP also used AI generation in most, maybe all, of their comment responses. The comments were engaging with the post and the user as though they were a human.
The last part is what surprised me, the HN crowd is as familiar with LLMs as any out there, how did they collectively fail to identify slop? Previously, AI generated prose I've seen posted on HN gets downvoted to death. Human judgement is really the only thing stopping online discussion from getting overwhelmed with slop.
For now, once you're familiar with AI generated text, it's pretty easy to spot. It's possible to get AI generated prose that looks more legitimate but that requires pretty comprehensive prompting/context strategies. I've been assuming that tech spaces, and most others, would nearly unanimously reject AI writing, it sucks to see that cracking at one of the most famous tech forums.
If people stop rejecting it, the Claws will happily take over content generation.
Perhaps the posters interacting are also mainly bots? A partially (and low certainty) testable hypothesis depending on how well known or how new the accounts are.
The thought occured to me, but I didn't care enough to check, they looked mostly legit. There was one undeniable bot reply that was downvoted to oblivion. Found the link
There's nothing stopping that from happening on Hacker News, and I find the culture promoted by HN to make dead-internet-operations even more likely there, than on Reddit.
I personally think that Hacker News users would be lulled into a false sense of security due to the fact that the average Hacker News post appears more articulate than on other, larger social media sites.
Honestly, I think the only reason why it appears that way is because there's a longstanding cultural aversion to "not being like Reddit" and a powerful downvote/flag feature that can outright hide posts, which does tend to happen to memes and low-effort replies.
But HN also has open registration, and access to moderation tools is only walled behind karma. The site already had a problem with voting rings and bad faith bots/alt accounts before AI made things ten times worse.
I think it's a pretty strong signal that a ton of comments from new posters are coming from LLMs. Whether they're from automated bots, or just illiterate morons doesn't really matter. If it looks like an LLM and smells like an LLM, in the absence of any better way to prove personhood, it should be treated as bot activity.
If people think it's unfair that their comments might get removed or their account suspended because they sound like a bot, then there's a really easy solution: write your own damn comments.
I have ADHD and I used to constantly use parenthesticals (for bonus content mid-sentence) but my girlfriend at the time—ex now but still a dear friend—convinced me that em-dashes were better looking. Now I use them out of habit and that coupled with my use of proper punctuation and capitalization has gotten me zero accusations of being AI. I think I must come across human enough to not need to prove myself. Probably when I use words like parenthetical… and my unhinged takes based on Vibes™
¿Porque no los dos? I think of a parenthetical as a brief, gentle aside, a slightly hushed tone you can switch into and back out of again without breaking the rhythm of the sentence. But an em-dash reads more like an abrupt self-interruption, like you’ve suddenly realized something mid-stride and have to inject it right here with some urgency.
Contrast those with an ellipsis, which feels to me more like trailing off, like you’re lost in thought and forgot where the sentence was going, so you just want to kinda leave it there and think for a sec, and maybe try to pick up the trail again in a moment. Or a semicolon, which is more versatile but I tend to use it like a period when what follows directly clarifies what came before it.
That’s how I interpret them anyway. Totally Vibes™ all the way down.
I use parentheses and em dashes to excess; also, semicolons. If I only stuck to one, my sentences would be unreadable (especially when — as demonstrated here — I start nesting them inside each other).
I'm not totally sure why I do this, but I suspect it's because I'm not a verbal thinker (no inner monologue, for example), and so I think in a more layered/non-linear way that's hard to translate cleanly into text.
I really don't like stylometric analyses offered up as proof of LLM usage, because LLMs aren't confined to a certain style, nor are humans incapable of using that style naturally themselves. The "telltale" signals might be em dashes today, and something completely different tomorrow. I'm especially concerned about those automated "AI checkers" marketed to teachers to catch cheating essay-writers, because they give the teachers a false sense of security... and their accuracy is seldom better than a coin toss, anyway.
Even so, this seems like a pretty strong indicator of bot activity, in this particular case at this moment in time. I do find it infuriating that bots are taking over more and more conversation places online, because they're most likely there to manipulate opinions for one purpose or another, and I don't want to waste my time having a good-faith discussion with disingenuous software.
Circling back on point 1. I love em dashes. I've used them for years. It's a totally natural part of my own writing style, as is being long-winded and sometimes throwing big words around from a soapbox. Not saying those are good writing habits... let's just call them identifiable ones. Sometimes I'm painfully aware of them and worry they make me sound awkward or stilted. I usually spend a good bit of time editing myself to file down the clunkiest parts, but ultimately I still am who I am and write how I write. And now with LLMs, a new fear has unlocked that people are going to start thinking I'm a bot just because my authorial voice overlaps with whatever ChatGPT's doing.
Oh jeez I just realized I made a three-point list... you see? YOU SEEEE??
The worst part about em-dashes isn't that AI uses them, it's that you're not supposed to put a space around it. Worst English style guide rule ever. /hj
But also, I absolutely agree with your last point. I'm a uni student and my part in a group paper recently got flagged as 100% AI by some software the teachers used. No one else's part was flagged. I guess I just write like AI, especially when using academic language. Being ESL definitely doesn't help either.
As part of my job, I have to localize technical documents between American English and Australian English. American English uses em dashes with no spaces—like this—whereas the style guide for our Australian documents use en dashes with spaces – like this – which I understand is also common in British writing.
So now I use a hybrid approach — em dashes with spaces — for my personal writing. Best of both worlds.
Yea I think people haven't really caught up to the idea that you should no longer assume a human is writing anything written on the internet starting in the last couple of years unless you have some idea of who that person is, digitally or otherwise. There is no fingerprint to any reasonably sophisticated use of AI engaging in account spoofing/astroturfing etc.
Which is kind of fine right now, because we all know people who post online from the before times. But in ~25 years? Yeesh...
fxgn | 6 hours ago
I think a more likely theory is that those accounts are not actual automated bots, it's just that HN used to mostly be for, well, hackers, but now more and more of the audience are AI startup bros who are also more likely to use AI to "phrase their thoughts" because they don't see anything wrong with that.
zod000 | 5 hours ago
This is probably the most likely theory. I can't stand reading HN anymore. It is so filled with AI Koolaid and startup tech bros that I find it unbearable. I mean, it has always had a high level of that sort, but it has both gotten much worse and my tolerance for it has evaporated.
JRandomHacker | 5 hours ago
Don't forget the rampant bigotry to go along with the AI-bro-ness!
tanglisha | 4 hours ago
They already said it's hacker news. That hasn't been a fun place for me in a very long time.
Nihilego | an hour ago
I haven’t used it for a long time but I had it an an RSS feed.
Every time I open it I feel like it’s just as astroturfed as some people claim reddit is, I stopped going on there after a while.
post_below | 5 hours ago
That matches what I've seen, one recent post that was at the top of the front page used clearly AI generated text to promote a (vibecoded) project. The OP also used AI generation in most, maybe all, of their comment responses. The comments were engaging with the post and the user as though they were a human.
The last part is what surprised me, the HN crowd is as familiar with LLMs as any out there, how did they collectively fail to identify slop? Previously, AI generated prose I've seen posted on HN gets downvoted to death. Human judgement is really the only thing stopping online discussion from getting overwhelmed with slop.
For now, once you're familiar with AI generated text, it's pretty easy to spot. It's possible to get AI generated prose that looks more legitimate but that requires pretty comprehensive prompting/context strategies. I've been assuming that tech spaces, and most others, would nearly unanimously reject AI writing, it sucks to see that cracking at one of the most famous tech forums.
If people stop rejecting it, the Claws will happily take over content generation.
Chiasmic | 5 hours ago
Perhaps the posters interacting are also mainly bots? A partially (and low certainty) testable hypothesis depending on how well known or how new the accounts are.
post_below | 4 hours ago
The thought occured to me, but I didn't care enough to check, they looked mostly legit. There was one undeniable bot reply that was downvoted to oblivion. Found the link
[OP] hungariantoast | 6 hours ago
This isn't proof of anything, but it is funny.
It's also easy to believe. Researchers have already demonstrated that they can deploy LLMs to:
There's nothing stopping that from happening on Hacker News, and I find the culture promoted by HN to make dead-internet-operations even more likely there, than on Reddit.
fefellama | 6 hours ago
Why do you find the culture on HN more susceptible to bots than Reddit? Not disagreeing, just curious.
Lexinonymous | 3 hours ago
I personally think that Hacker News users would be lulled into a false sense of security due to the fact that the average Hacker News post appears more articulate than on other, larger social media sites.
Honestly, I think the only reason why it appears that way is because there's a longstanding cultural aversion to "not being like Reddit" and a powerful downvote/flag feature that can outright hide posts, which does tend to happen to memes and low-effort replies.
But HN also has open registration, and access to moderation tools is only walled behind karma. The site already had a problem with voting rings and bad faith bots/alt accounts before AI made things ten times worse.
babypuncher | 3 hours ago
I think it's a pretty strong signal that a ton of comments from new posters are coming from LLMs. Whether they're from automated bots, or just illiterate morons doesn't really matter. If it looks like an LLM and smells like an LLM, in the absence of any better way to prove personhood, it should be treated as bot activity.
If people think it's unfair that their comments might get removed or their account suspended because they sound like a bot, then there's a really easy solution: write your own damn comments.
Narry | 3 hours ago
I have ADHD and I used to constantly use parenthesticals (for bonus content mid-sentence) but my girlfriend at the time—ex now but still a dear friend—convinced me that em-dashes were better looking. Now I use them out of habit and that coupled with my use of proper punctuation and capitalization has gotten me zero accusations of being AI. I think I must come across human enough to not need to prove myself. Probably when I use words like parenthetical… and my unhinged takes based on Vibes™
hamstergeddon | 2 hours ago
Damn just learned a new cheat code for ADHD—albeit now I appear to be AI rather than insane.
balooga | an hour ago
¿Porque no los dos? I think of a parenthetical as a brief, gentle aside, a slightly hushed tone you can switch into and back out of again without breaking the rhythm of the sentence. But an em-dash reads more like an abrupt self-interruption, like you’ve suddenly realized something mid-stride and have to inject it right here with some urgency.
Contrast those with an ellipsis, which feels to me more like trailing off, like you’re lost in thought and forgot where the sentence was going, so you just want to kinda leave it there and think for a sec, and maybe try to pick up the trail again in a moment. Or a semicolon, which is more versatile but I tend to use it like a period when what follows directly clarifies what came before it.
That’s how I interpret them anyway. Totally Vibes™ all the way down.
RoyalHenOil | an hour ago
I use parentheses and em dashes to excess; also, semicolons. If I only stuck to one, my sentences would be unreadable (especially when — as demonstrated here — I start nesting them inside each other).
I'm not totally sure why I do this, but I suspect it's because I'm not a verbal thinker (no inner monologue, for example), and so I think in a more layered/non-linear way that's hard to translate cleanly into text.
balooga | 5 hours ago
My thoughts are threefold:
I really don't like stylometric analyses offered up as proof of LLM usage, because LLMs aren't confined to a certain style, nor are humans incapable of using that style naturally themselves. The "telltale" signals might be em dashes today, and something completely different tomorrow. I'm especially concerned about those automated "AI checkers" marketed to teachers to catch cheating essay-writers, because they give the teachers a false sense of security... and their accuracy is seldom better than a coin toss, anyway.
Even so, this seems like a pretty strong indicator of bot activity, in this particular case at this moment in time. I do find it infuriating that bots are taking over more and more conversation places online, because they're most likely there to manipulate opinions for one purpose or another, and I don't want to waste my time having a good-faith discussion with disingenuous software.
Circling back on point 1. I love em dashes. I've used them for years. It's a totally natural part of my own writing style, as is being long-winded and sometimes throwing big words around from a soapbox. Not saying those are good writing habits... let's just call them identifiable ones. Sometimes I'm painfully aware of them and worry they make me sound awkward or stilted. I usually spend a good bit of time editing myself to file down the clunkiest parts, but ultimately I still am who I am and write how I write. And now with LLMs, a new fear has unlocked that people are going to start thinking I'm a bot just because my authorial voice overlaps with whatever ChatGPT's doing.
Oh jeez I just realized I made a three-point list... you see? YOU SEEEE??
fxgn | 4 hours ago
The worst part about em-dashes isn't that AI uses them, it's that you're not supposed to put a space around it. Worst English style guide rule ever. /hj
But also, I absolutely agree with your last point. I'm a uni student and my part in a group paper recently got flagged as 100% AI by some software the teachers used. No one else's part was flagged. I guess I just write like AI, especially when using academic language. Being ESL definitely doesn't help either.
RoyalHenOil | 57 minutes ago
As part of my job, I have to localize technical documents between American English and Australian English. American English uses em dashes with no spaces—like this—whereas the style guide for our Australian documents use en dashes with spaces – like this – which I understand is also common in British writing.
So now I use a hybrid approach — em dashes with spaces — for my personal writing. Best of both worlds.
Chiasmic | 5 hours ago
At least there isn’t an emoji every other sentence
balooga | 4 hours ago
💯 You’re absolutely right! 😅
JCPhoenix | 3 hours ago
👀
🤔
🤖
streblo | 4 hours ago
Yea I think people haven't really caught up to the idea that you should no longer assume a human is writing anything written on the internet starting in the last couple of years unless you have some idea of who that person is, digitally or otherwise. There is no fingerprint to any reasonably sophisticated use of AI engaging in account spoofing/astroturfing etc.
Which is kind of fine right now, because we all know people who post online from the before times. But in ~25 years? Yeesh...