The problem is on a thread where there is 1 AI post, people will identify 10 posts as being AI.
People latch onto the word "delve" or an em-dash or the idiom of "it's not X it's Y" as being proof that something was written by AI, without ever considering that AI is largely just writing in the style of how internet commenters write since that's a lot of the training data.
Now we've got students re-writing their homework to avoid looking like AI and commenters self-censoring against words and phrases that trigger AI suspicion.
I'm deeply skeptical that people can correctly identify human-written vs latest AI model written HN comments with sufficient accuracy.
It doesn't seem to work all that well, I still see posts about AI. Maybe it should read the text content then use a classifier to see whether it's AI related or not, basic LLMs work pretty well at that and you honestly could just use a classical AI classifier like Naive Bayes filter too.
Joking or not, in fact neural networks as classifiers are obviously useful: detecting cancer in medical images, for example. And that approach would, I’m sure, work better here than my crude regex. But this use of AI is distinct from its use as a giant plagiariser and slop generator. What I and others are tired of seeing are endless stories about vibe coded games and whatnot.
I’ve been looking at taking a step back from the internet as a whole, including HN, because I’m so tired of the AI slop.
HN and YouTube are basically the only sites I still use. I disabled my watch history on YouTube to kill the recommendations page to get away from a lot of the trash and just focus on my subscriptions. I’ll be unsubscribing to those writing scripts with AI as well, as I notice them.
I’d like to be able to flag things as AI slop on YouTube, so the person gets that feedback and can hopefully get back on the right track. On here, I’m not so sure if that type of reinforcement will matter, or if will be more obvious to the person why a post was flagged.
I read they are supposed to be labeling AI stuff, but I assume this is with their tool for looking for the fingerprinting they add to AI generated content. I don’t expect this to help with actual people reading an AI generated script.
I don’t know that they care, as I still see a lot of engagement on these videos, where only one person will call a video out as AI and not even get many upvotes for it. It takes a lot of scrolling to find it. I expect YouTube might say what is AI, but not have a global filter to remove it… but I hope they prove me wrong.
This is only an idea, but how about a Summarize button that does use AI to write a very concise blurb of the comment. Then if it looks stupid enough at first glance, just read the summary and move on.
We already have a problem with people refusing to engage with the content and only commenting on comments or the title. The last thing we need is to discourage people from reading the article even more than they already don't.
> Users flagged the post as breaking the guidelines or otherwise not belonging on HN.
I'd favor having a second type of tag, for submissions, which meant "the linked article is of low quality". Doesn't matter to me whether it's AI slop, or press release puffery, or tedious drivel, or by a painfully unqualified author, or something else.
I wonder if the signal people actually want is "low information density" rather than "AI-generated."
A lot of the frustration seems to come from content that takes 2,000 words to say something that could have been said in 200, regardless of whether a human or a model wrote it.
If a post is original, useful, and teaches me something, I don't care much how it was produced. What I notice is when a lot of words are used to communicate very little.
Yes, but low information quality is even worse than low information density. If I happen to know the subject well, and the article contains glaring errors, obvious omissions, or miserable fudges - I'd like a quick way to tag it as dubious.
The problem is on a thread where there is 1 AI post, people will identify 10 posts as being AI.
People latch onto the word "delve" or an em-dash or the idiom of "it's not X it's Y" as being proof that something was written by AI, without ever considering that AI is largely just writing in the style of how internet commenters write since that's a lot of the training data.
Now we've got students re-writing their homework to avoid looking like AI and commenters self-censoring against words and phrases that trigger AI suspicion.
I'm deeply skeptical that people can correctly identify human-written vs latest AI model written HN comments with sufficient accuracy.
Doesn't the flag button serve that purpose?
I don't think it does exactly. I edited to clarify.
Of course it does. If it’s slop it should be flagged (I say this after reading the edited submission).
In the meantime you can avoid at least some of the slop by using https://hn-ai.org/ or one of the other anti-slop extensions or alternative sites.
It doesn't seem to work all that well, I still see posts about AI. Maybe it should read the text content then use a classifier to see whether it's AI related or not, basic LLMs work pretty well at that and you honestly could just use a classical AI classifier like Naive Bayes filter too.
Yes, there are still plenty of AI stories getting past my filter. It’s just a regex on the titles.
You should use an LLM.
Joking or not, in fact neural networks as classifiers are obviously useful: detecting cancer in medical images, for example. And that approach would, I’m sure, work better here than my crude regex. But this use of AI is distinct from its use as a giant plagiariser and slop generator. What I and others are tired of seeing are endless stories about vibe coded games and whatnot.
Down voting submissions would be a better feature, so much slop gets posted here as gospel until someone realizes its slop generated.
I’ve been looking at taking a step back from the internet as a whole, including HN, because I’m so tired of the AI slop.
HN and YouTube are basically the only sites I still use. I disabled my watch history on YouTube to kill the recommendations page to get away from a lot of the trash and just focus on my subscriptions. I’ll be unsubscribing to those writing scripts with AI as well, as I notice them.
I’d like to be able to flag things as AI slop on YouTube, so the person gets that feedback and can hopefully get back on the right track. On here, I’m not so sure if that type of reinforcement will matter, or if will be more obvious to the person why a post was flagged.
https://unhook.app/ lets you disable features on YT (granular though the default is good).
https://aisloplist.com/ works as well but considering the torrent of AI shite posted on YT it only gets some of it but better than nothing.
I wish there would be a mark on YouTube AI slop or a way to mark AI generated garbage.
Its is likely coming, but needed yesterday.
I read they are supposed to be labeling AI stuff, but I assume this is with their tool for looking for the fingerprinting they add to AI generated content. I don’t expect this to help with actual people reading an AI generated script.
I don’t know that they care, as I still see a lot of engagement on these videos, where only one person will call a video out as AI and not even get many upvotes for it. It takes a lot of scrolling to find it. I expect YouTube might say what is AI, but not have a global filter to remove it… but I hope they prove me wrong.
This is only an idea, but how about a Summarize button that does use AI to write a very concise blurb of the comment. Then if it looks stupid enough at first glance, just read the summary and move on.
We already have a problem with people refusing to engage with the content and only commenting on comments or the title. The last thing we need is to discourage people from reading the article even more than they already don't.
What if the submission said the most common reason an article was flagged next to the title?
That way if it's AI slop it'll say [AI slop], if it's spam it'll say [Spam], if it's dubiously legal/illegal content it'll say that, etc?
You could use that to decide if you want to give the submission a chance or not.
Too Many Slops and you can't post for a week. Yes.
Just one slop would be sufficient.
If we wanna know ai opinion we all can get it.
From HN's FAQ:
> What does [flagged] mean?
> Users flagged the post as breaking the guidelines or otherwise not belonging on HN.
I'd favor having a second type of tag, for submissions, which meant "the linked article is of low quality". Doesn't matter to me whether it's AI slop, or press release puffery, or tedious drivel, or by a painfully unqualified author, or something else.
I wonder if the signal people actually want is "low information density" rather than "AI-generated."
A lot of the frustration seems to come from content that takes 2,000 words to say something that could have been said in 200, regardless of whether a human or a model wrote it.
If a post is original, useful, and teaches me something, I don't care much how it was produced. What I notice is when a lot of words are used to communicate very little.
Yes, but low information quality is even worse than low information density. If I happen to know the subject well, and the article contains glaring errors, obvious omissions, or miserable fudges - I'd like a quick way to tag it as dubious.
If you really know it’s dubious post a comment.