I can deal with a launch problem. I am looking forward to being able to check it out. I stopped coming to HN as often as I used to because there has been too much AI talk. It's like a dang AI subreddit. I don't want anything to do with AI. I have lost 2 jobs to AI budgets and stupid executive decisions. It's no longer part of my personal and professional life.
I agree with you, but haven't really left yet since there's still a good amount of discourse outside AI related topics. Seems a significant amount of contributors here are all to eager to fall over themselves putting themselves, their juniors, children, and colleagues out of work, long term.
My personal stance is that the genie is out of the bottle, and there's likely no putting it back. Kind of like nuclear weapons: I wish they didn't exist, but since they do I'd rather be on the side that has them.
Honestly it's an incredible technology the likes of which I never thought I would see in my lifetime. It's definitely not perfect, but it is improving at a frightening pace. Some days I am optimistic for how it might shape the future, and other days I am scared.
If goverments wanted to make a credible effort to limit how they are used, I would support that. But I'm not terribly optimistic that can or will happen.
It’s interesting how, as someone more pro-AI, I feel the exact opposite way - every comment thread just devolves into people saying AI is stupid (OK, perhaps modulo new model releases from frontier labs, those tend to be fairly positive). Like we had a thread about what your AI workflow is and people were coming in and saying they didn’t have one because AI is bad. Really? Do these people enter threads about what your favorite Rust library is and say they don’t have any and Rust is a bad language?
Sure, but not to the same extent. Because the industry isn't experiencing a trend where it tells everyone that if they don't use Rust, they are obsolete. And because nobody is saying "Rust is going to take all our jobs" to all the other programmers. And because most of the threads on HN aren't about Rust on a daily basis.
Sure overly enthusiastic Rust fans exist and they are annoying. But it's nowhere even close to the AI mania gripping the industry.
Just because some people are contributing to bad discourse doesn't mean that you also get permission to contribute to bad discourse. If you don't like it when people say you're going to be obsolete -- tell them! Don't hop into random spaces where no one is saying that and act negative! This just lowers the discourse quality everywhere. Very few people on HN are saying you are going to be obsolete if you don't use AI (I basically haven't seen a single person say this who hasn't been massively downvoted).
> Do these people enter threads about what your favorite Rust library is and say they don’t have any and Rust is a bad language?
Yes. And Rust people enter threads about other languages and projects not written in Rust and say they're bad because they aren't Rust. Welcome to Hacker News.
The style is a bit off for the usual HN fare: 'sans' should be lowercase here, or otherwise it's easy to confuse for a font. It's still pretty swell :)
The only way to fully detect and remove AI posts from HN is using human curation by a human-selected few. Everything automated will have a higher review rate (postsReviewed / postsTotal) and also an higher error rate (incorrectReview / postsReviewed) in return for not being subject to puny human weaknesses like sleep, sickness, and (until curator selection is itself automated), spam/troll malice. To misquote my favorite blooper reel ever, “That’s not something that code can fix, that’s gonna be a little harder to fix”.
I built (well, vibe-coded) a version of this a while back that runs against the Hacker News API, it's static HTML on GitHub Pages so I don't have to run a server for it: https://tools.simonwillison.net/hacker-news-filtered
Immediately I started thinking how nice it would be to use natural language to have LLMs generate a deterministic filter for stories matching content I DO care about, filtered from New. Instead of filtering it out.
I have an app that does Hacker News with AI; it analyses all the stories and comments for a number of criteria, and tags them so you can skip stuff you don't want to see. I should get around to actually publishing it but I've been lazy.
One of the fun things I noticed is the psychological impact of framing. A comment that might've made you feel the need to reply before has less emotional weight if it's highlighted in red and a diminished font. Same thing for stories; if you would normally disagree with a story and it would make you want to comment, you feel less like commenting if the story is rated as 'lacking evidence', 'unsupported by research', 'personal anecdotes only', etc. It drives down the feeling of needing to engage. Which is horrible for site engagement, but good for mental health (I think).
73% of users vote prior to reading TFA, according to this research. (I am sometimes guilty of this myself)
We live in a world being dimished by confirmation bias, but this isn't a new thing. Those who wrote/approved the headlines always had more power than those who wrote the articles.
> In the present work, we introduce and make available a new dataset containing the activity logs that recorded all activity for 309 Reddit users for one year.
However, tracking over a year might make the subject forget about the whole thing, and act naturally. As far as HN vs Reddit, not much difference really. I mean that as more props to reddit users than anything against HNers.
I got a partial load and what it looks like it does is just search each submission for a list of key words and discards any that hits, so it would discard this submission.
I was not suggesting that it would be a false positive, I was suggesting that this will filter out many submissions that would be of interest to those that want less AI on HN. This would flag a blog that has nothing to do with AI if some random person mentioned AI in the comments of that blog post, right?
I've been thinking of making something like this for myself for quite a while now, glad I'm not the only one who had the idea and someone actually did it before I got to it
Great idea. But the first thread I opened (Nordstjernen 1.0) is according to the comments about a project built with Claude. So it doesn't remove all the AI stuff.
What would be the worse non-semantic alternative? Regex?
LLMs are the holy grail for getting beyond string matches. I would hope one was used to solve such a problem, otherwise that would be a poor product, right?
I did not investigate the product, but my point here is irony. The correct solution to implement the TFA product is to use an LLM.
Yes, keyword-based matching is more than good enough for the use case. It doesn't justify bringing in trillion dollar companies to subsidize the costs.
Even if you wanted to go beyond simple keyword matching, there are ways to determine relevance before jumping to the shiny new overkill solution.
Also for everyone's sanity, please don't turn a troll's false accusations into a cynical argument for more LLMs. That's what drives to people to reach for solutions like the one in the article in the first place.
I built a filter for AI (inclusive and exclusive) for hcker.news a while back and continue to maintain it. It’s quite robust, and does more than just keyword filtering.
I have my own version of this as a browser extension paired with a backend that runs all new submissions through a small LLM to classify them, which catches more than a simple word match. Fight fire with fire, as they say.
Though I haven't used it much lately, because seeing half of the front page disappear when I enable it is a bit disheartening.
I can deal with a launch problem. I am looking forward to being able to check it out. I stopped coming to HN as often as I used to because there has been too much AI talk. It's like a dang AI subreddit. I don't want anything to do with AI. I have lost 2 jobs to AI budgets and stupid executive decisions. It's no longer part of my personal and professional life.
I hope it works out.
I agree with you, but haven't really left yet since there's still a good amount of discourse outside AI related topics. Seems a significant amount of contributors here are all to eager to fall over themselves putting themselves, their juniors, children, and colleagues out of work, long term.
My personal stance is that the genie is out of the bottle, and there's likely no putting it back. Kind of like nuclear weapons: I wish they didn't exist, but since they do I'd rather be on the side that has them.
Honestly it's an incredible technology the likes of which I never thought I would see in my lifetime. It's definitely not perfect, but it is improving at a frightening pace. Some days I am optimistic for how it might shape the future, and other days I am scared.
If goverments wanted to make a credible effort to limit how they are used, I would support that. But I'm not terribly optimistic that can or will happen.
> Kind of like nuclear weapons
True, but you're not being bombarded 24/7 with irrelevant -- and most likely false -- information about nuclear weapons.
How a post about "Hacker News, Sans AI" became about .... AI !
Distinction is perfect continence.
TBH, I'm mostly here out of spite at this point. It's unhealthy but I can't seem to stop rage-reading.
Would love to hear more detail about the job losses you mentioned, if you are willing to share.
I've addressed the issues. It should be usable now. I apologize to everyone who encountered an error page. Have a good weekend, everyone!
It’s interesting how, as someone more pro-AI, I feel the exact opposite way - every comment thread just devolves into people saying AI is stupid (OK, perhaps modulo new model releases from frontier labs, those tend to be fairly positive). Like we had a thread about what your AI workflow is and people were coming in and saying they didn’t have one because AI is bad. Really? Do these people enter threads about what your favorite Rust library is and say they don’t have any and Rust is a bad language?
Sure, but not to the same extent. Because the industry isn't experiencing a trend where it tells everyone that if they don't use Rust, they are obsolete. And because nobody is saying "Rust is going to take all our jobs" to all the other programmers. And because most of the threads on HN aren't about Rust on a daily basis.
Sure overly enthusiastic Rust fans exist and they are annoying. But it's nowhere even close to the AI mania gripping the industry.
Just because some people are contributing to bad discourse doesn't mean that you also get permission to contribute to bad discourse. If you don't like it when people say you're going to be obsolete -- tell them! Don't hop into random spaces where no one is saying that and act negative! This just lowers the discourse quality everywhere. Very few people on HN are saying you are going to be obsolete if you don't use AI (I basically haven't seen a single person say this who hasn't been massively downvoted).
> Do these people enter threads about what your favorite Rust library is and say they don’t have any and Rust is a bad language?
Yes. And Rust people enter threads about other languages and projects not written in Rust and say they're bad because they aren't Rust. Welcome to Hacker News.
If you can find me even a single example of this I’d love to see it.
The rsync thread has the requisite comment saying a Rust rewrite would fix everything.
I really miss n-gate
I've been training to naturally ignore AI mentions that I thought this entry was about some HN ui font change (Sans).
The style is a bit off for the usual HN fare: 'sans' should be lowercase here, or otherwise it's easy to confuse for a font. It's still pretty swell :)
> The style is a bit off
A much better option, in my opinion: https://hcker.news/?ai=exclude
Yeah, that doesn't work either. E.g. it has the following stories:
- Nordstjernen 1.0 (built with Claude)
- Ask HN: What was your "oh shit" moment with GenAI?
The only way to fully detect and remove AI posts from HN is using human curation by a human-selected few. Everything automated will have a higher review rate (postsReviewed / postsTotal) and also an higher error rate (incorrectReview / postsReviewed) in return for not being subject to puny human weaknesses like sleep, sickness, and (until curator selection is itself automated), spam/troll malice. To misquote my favorite blooper reel ever, “That’s not something that code can fix, that’s gonna be a little harder to fix”.
I don't see the "Claude" story. Did you change the filters? (i can clearly see a message that the AI filter is active in the top of the page.)
"GenAI" is not filtered indeed. Maybe it's not working perfectly, but it does work for most/many AI stories. Good enough!
PS. You can just add the word "GenAI" to your manual filters, in settings...
Was about to write the same
I built (well, vibe-coded) a version of this a while back that runs against the Hacker News API, it's static HTML on GitHub Pages so I don't have to run a server for it: https://tools.simonwillison.net/hacker-news-filtered
It is a little ironic coming from the most prolific AI poster here
I have a quirky sense of humor.
interesting but i rarely use any HN reader
there's just something about this UI and its consistency
I also don't mind all the AI related news
If anything I just wish they had a mute/block button. its not fun when somebody is stalking you replying to every comment you make.
How often do you use it?
Immediately I started thinking how nice it would be to use natural language to have LLMs generate a deterministic filter for stories matching content I DO care about, filtered from New. Instead of filtering it out.
I don't use it myself. I built it for a Hacker News thread where someone else was saying "I wish I could browse Hacker News without all the AI stuff".
Love that this is configurable unlike OPs version. That way I can exclude "I rewrote ____ in Rust".
Would be nice to have "include" option... for those of use who may want to ONLY read posts about AI or whatever else niche interest.
I have an app that does Hacker News with AI; it analyses all the stories and comments for a number of criteria, and tags them so you can skip stuff you don't want to see. I should get around to actually publishing it but I've been lazy.
One of the fun things I noticed is the psychological impact of framing. A comment that might've made you feel the need to reply before has less emotional weight if it's highlighted in red and a diminished font. Same thing for stories; if you would normally disagree with a story and it would make you want to comment, you feel less like commenting if the story is rated as 'lacking evidence', 'unsupported by research', 'personal anecdotes only', etc. It drives down the feeling of needing to engage. Which is horrible for site engagement, but good for mental health (I think).
For one sec, I got excited for a new font!
Interesting this keeps moving up the front page when the site is inaccessible because it's hosted on a baked potato.
Yep. My single core Proxmox machine sitting in my living room with horrible cache invalidation is struggling.
Sounds like a great use case for serving plain HTTP over cloudflare tunnel. Why terminate TLS when you don't have to?
You know what's funny? Most of this site is exactly that. The _one_ time I do something dynamic…
73% of users vote prior to reading TFA, according to this research. (I am sometimes guilty of this myself)
We live in a world being dimished by confirmation bias, but this isn't a new thing. Those who wrote/approved the headlines always had more power than those who wrote the articles.
https://www.researchgate.net/publication/315096490_Consumers...
edit: disclaimer, no hate on TFA. Just responding to the comment.
73% of Reddit users vote prior to reading TFA:
> In the present work, we introduce and make available a new dataset containing the activity logs that recorded all activity for 309 Reddit users for one year.
n=309 is better than n=1, but yeah...
However, tracking over a year might make the subject forget about the whole thing, and act naturally. As far as HN vs Reddit, not much difference really. I mean that as more props to reddit users than anything against HNers.
Reddit "users"
Huh, I didn't even remember to vote for posts, i just scroll down to read (and vote sometimes on comments)
This submission should be studied by people holding clipboards. It needs a follow-up:
"How I made it to the top of HN with zero content beyond a catchy title"
It further proves the key to getting your stuff on HN is not to post interesting content, it's to post something that sounds interesting.
D1 hater right here.
Shows that there is demand for this. It works for me now btw.
Just goes to show how much demand there is.
HN really needs a containment board.
I'm curious to find out if they're using a model to detect AI content. I'd be more amused than disappointed. But, alas, potato.
I got a partial load and what it looks like it does is just search each submission for a list of key words and discards any that hits, so it would discard this submission.
I would argue that in a small way, this post _is_ about AI, so it wouldn't be a false positive.
I was not suggesting that it would be a false positive, I was suggesting that this will filter out many submissions that would be of interest to those that want less AI on HN. This would flag a blog that has nothing to do with AI if some random person mentioned AI in the comments of that blog post, right?
It's not an LLM.
Why use big program when regex do trick?
You can get the best of both worlds with a small embedding model
Show me an embedding model that runs as fast as an optimized regex engine and I'll buy you a beer.
Probably not as fast as a simple regex but static embedding models can get stupid fast e.g https://www.flowercomputer.com/news/fast-static-embedding/
Master of efficiency
It should be accessible now.
My immediate thought was already expressed in one of the few comments:
> Does this use an LLM to categorize "AI-related" vs "not-AI related" articles? Would be ironic. Lol
I've been thinking of making something like this for myself for quite a while now, glad I'm not the only one who had the idea and someone actually did it before I got to it
Was hoping it was a new font.
The filter doesn't appear to be perfect, one of the top posts right now is "My Agent Skill for Test-Driven Development"
I guess they could run the titles through a (flash) LLM to catch ones were keyword detection won't cut it.
Have to love the irony. :)
At least the blog post was proofread by an AI.
That should be fixed as soon as my server has a chance to catch up.
Great idea. But the first thread I opened (Nordstjernen 1.0) is according to the comments about a project built with Claude. So it doesn't remove all the AI stuff.
Hug of death?
yep. standby
A lot of demand for ‘Sans AI’ :)
Probably by bots :)
Our AI overlords don't want an AI-free Hacker News
Oh. A little bit of a let down. I was expecting "HN without comments and posts written by LLMs".
That would be cool. Although your username is a little suspicious…
Are you using AI to figure out what's about what's AI and what's not?
Hopefully fixed the cache invalidation. I'm crossing my fingers.
Coincidently he uses AI to solve this problem
Do you have evidence? I see no indication of this.
I have evidence that it's not AI. It's just a regex.
https://github.com/elijah-potter/blog/blob/master/pages/hnsa...
I thought as much. With the sheer amount of flagged comments in the comment history turning up with showdead on, it looked very suspicious.
What would be the worse non-semantic alternative? Regex?
LLMs are the holy grail for getting beyond string matches. I would hope one was used to solve such a problem, otherwise that would be a poor product, right?
I did not investigate the product, but my point here is irony. The correct solution to implement the TFA product is to use an LLM.
Yes, keyword-based matching is more than good enough for the use case. It doesn't justify bringing in trillion dollar companies to subsidize the costs.
Even if you wanted to go beyond simple keyword matching, there are ways to determine relevance before jumping to the shiny new overkill solution.
Also for everyone's sanity, please don't turn a troll's false accusations into a cynical argument for more LLMs. That's what drives to people to reach for solutions like the one in the article in the first place.
Sometimes a little poison is medicine.
While you are at it can you filter out posts to subscription only sites eg economist NYT $&c. We're all here to escape click bait
A problem is it doesn't load as fast. Could it be helped?
Yes, it can. I am having a major cache invalidation problem. One of the major blights of programmers everywhere.
Great idea. I too have been worn out by AI talk on HN.
Right now, number 3 is an AI slop tutorial released by Microsoft.
I guess there is no real escape
^how about just be a devoted AI section at the top?
Welp, it was only a matter of time. Thank you
Yes, please. A general-purpose filter like the one on 4chan would also work.
I built a filter for AI (inclusive and exclusive) for hcker.news a while back and continue to maintain it. It’s quite robust, and does more than just keyword filtering.
https://hcker.news/?ai=exclude
Someone mentioned this in another comment, so I'll just copy/pasta my own comment:
Yeah, that doesn't work either. E.g. it has the following stories:
- Nordstjernen 1.0 (built with Claude)
- Ask HN: What was your "oh shit" moment with GenAI?
Thanks, love it!
I'm now wondering, what the default filter of https://hckrnews.com/ is exactly. I can't seem to get the same result on https://hcker.news/ ....
:/
...so i think its just a cumulative view of the Hacker News front page, except grouped/ordered differently? ¯\_(ツ)_/
I have my own version of this as a browser extension paired with a backend that runs all new submissions through a small LLM to classify them, which catches more than a simple word match. Fight fire with fire, as they say.
Though I haven't used it much lately, because seeing half of the front page disappear when I enable it is a bit disheartening.
Maybe HN could have a filter button that hides ANY threads for keywords you choose to filter.
I would instantly hide anything about Elon Musk or Jeff Bezos.
Says the guy that uses a man-in-the-middle service to "protect" his website. So dumb. I seriously hate Cloudflare.
I've been thinking the same. Couldn't this be done with a browser extension that hides elements that matches a regex?
Not quite. Lots of articles don't mention AI in the title, even if that is their core subject.
Okay, now filter out the VC stuff too...
Love this. Really nice to see a HN front page without AI, the fatigue is real.