You're underselling it. It's less than in their second month of existence - and that has been true every month for the past year. For the last few months it's also lower than their first month. Stack Overflow, effectively, no longer exists.
And we know why. It's because of their hostile attitude towards their users.
No doubt they'll continue operating until there's no more value in selling AI data licenses, then they'll unplug the servers and unregister the business.
Well, the community never made it a very welcoming place... But, I think a part of this is that it's just faster to throw a Q into ChatGPT. There's no waiting for a response, a major bonus is the non-hostility + willingness to respond to 'stupid' questions.
> And we know why. It's because of their hostile attitude towards their users.
I used SO almost every day until capable LLMs showed up. I don't remember the last time I used it. SO and similar sites are dead, and LLMs killed them.
> And we know why. It's because of their hostile attitude towards their users.
Uh, no.
StackOverflow is and has long been one of the most hostile places on the web, I’m not denying that. But this unprecedented drop in activity is 100% because of AI. Their product is as obsolete as horse-drawn carriages.
It was dying well before AI and while a knee happened coincidentally to chatgpt release, I don't think you remember correctly what gpt answer looked like for programmers questions back then. It was much later where it started assisting development.
I don't understand the motivation to reattribute the death of SO to anything other than AI, which it unquestionably was. GPT-3.5 was able to answer some small chunk of my programming questions (like 10-20% of them), which seems clearly reflected in the graph.
It doesn't matter. Even if it were the most friendly community ever, as long as their core value were the same - answering programming questions that with clear answers - it would have been replaced by today's AI completely. It's irrelevant today how bad GPT3.5 once was.
I miss Server Fault-- more the time in my life, I suppose, than the site itself.
I enjoyed the hell out of answering questions on Server Fault for a few years. It came along around the time my teaching gig at the local community college transitioned to online-only classes (which, oddly enough, I couldn't find enthusiasm to teach). I feel like the Server Fault community tried to make it a more hospitable place than Stack Overflow, FWIW, back in the day.
When Stack Exchange added the realtime chat system I got into interacting with members of the community more than could be done in questions or comments. There were even a couple people who ended up being IRL acquaintances. There were also wacky hijinks, too: https://meta.serverfault.com/a/1854 (I'm agog that those things happened 15 years ago, too... Yikes!)
Then we had our daughter and my time to dedicate to the site became scarce. I knew it was in the cards for me to lose the top spot on the leaderboard so I decided to just pack it in and quit cold turkey. I think I'm still fairly high on the leaderboard but, other than occasionally getting search engine results that point me to the site, I've stayed away. I don't live there anymore, as it were.
Edit: My favorite part of the site was having friends and Customers tell me about times they'd search-engined a question and found themselves using one of my answers to solve their problem. At least one Customer was amused to discover I'd fictionalized their particular situation as an example in an answer.
1) humans (AI assisted or not) answering with real-money stakes, and
2) people betting on which humans are correct (borrow page from prediction markets)
This does four things simultaneously:
1) It introduces consequences for being wrong, which in turn introduces real asymmetric incentives to be contrarian-but-correct
2) It creates real-world incentives for human guided votes to answers (because, again, there's a real-world stake)
3) Together, this gets humans back on the platform answering questions, which creates meaningful exhaust data that can be sold back to the AI companies.
4) It creates incentives to game the system, which creates yet-more exhaust data for training data for anti-fraud AI systems
I know many people who have got banned and never managed to get back on Reddit because Reddit is using some very invasive and persistent banned user detection techniques.
Thing is, they use AI to ban people; over time, many false positives accumulate and not everyone has the time and energy to appeal bans.
Slowly, more and more user accounts get banned, and even technical people I know fail to make a new account which lasts more than 3-4 days.
But the Reddit management is seeing this as a "win," not able to see the consequence of doing this.
There was that 1 person who used to work(paid?/volunteer?) at Reddit that actively tried to cultivate high-quality AMA on Reddit. Ever since they've been gone. It hasn't felt the same.
Some amount of hope and optimism that was around has definitely dried up or at least moved elsewhere into other smaller communities.
Everything is overly centralized around a particular influencer/podcaster/celebrity/product or whatever nowadays, it's harder to find topic based or open community-based places. The other places still exist, but still difficult to find and keep going.
I do hope the pendulum swings back towards hopeful optimistic culturally rich community vibes in many digital spaces.
I stopped using Reddit not because I got banned, but because it has been steadily converted from the world's most popular forum platform to a trash heap of engagement hacks and shady practices. Moreso than SO, which was already struggling with existential problems before AI, Reddit did itself in.
My account of well over a year and 11k karma was banned by admins presumably for an admin-deleted comment where I said "Hi kids, buy our product!" as a Steve-Buschemi-meme reference, and was flagged as spam. To your point they also have their Comment Quality Score (CQS) which lowers you automatically if you use Proton email aliases, lowers you if you use VPN to access the site, among other things.
Made several polite appeals and requests for clarification, all which went unanswered.
I refuse to make a Reddit account. When they killed "old.reddit.com" for non-logged-in users I stopped reading it at all. It's completely useless to me now.
I used to be a lot more enthusiastic about decentralized protocols. I don't know how to solve the AI problem though. I think the Internet as we knew it for assumed-human interaction is dead, and the surest way to believe you're interacting with humans now is through existing trusts, such as accounts older than AI and highly-moderated forums, and to charge for access.
I didn't do much community stuff on Stack Overflow, but Server Fault suffered from a certain class of user who was more interested in curating and rules-lawyering the "quality" of questions than actually helping people. It was hard work to keep those people in check, and the Stack Exchange metaphor seemed to draw those kind of people like moth to flame.
SO wasn't a discussion forum, it was wikipedia for programming. It was already dying long before the question rate dropped off, since most new questions were "plz write/debug my code for me" which tend to be the sort of question that SO was not designed for. Chatgpt replaced that, and github issues replaced project-specific debugging. Now there's nobody left to write/read/answer the small amount of questions that are still a good fit.
I really do love their other communities so much more. Lots of super interesting content - it feels like a new digg or something. I go to one of the other stacks and look at the most popular non-tech stack posts once or twice a week purely for entertainment. The breakdowns on obscure but interesting topics can get really fun to read!
On the more-and-more rare occasion that I end up on Stack Overflow as a result of a search I find myself losing excessive time to following the random links to other communities. The other communities really are good stuff.
The golden age of SO was when all the important questions that have accumulated over decades were getting asked and answered. SO stabilized once that backlog got finished, and then started dying once people moved to github issues for project-specific questions. Chatgpt just replaced the last remaining type of questions ("plz do my work for me"), which arguably shouldn't have been getting asked in the first place.
The peak number of questions was in 2013, and they had roughly zero growth between 2013 & 2016. They were in decline long before LLMs were on the scene.
If there is any proof that no one will be writing code, this is it.
I encourage all young developers to focus on information architecture, application architecture, pure logic, domain-driven design, and harness engineering.
I have some wishful thinking for everyone reading this: what if the people at StackOverflow built one of the best bot protections out there? I would honestly expect the graph to go up with all the LLMs and automated content spamming..
I answered a number of questions on there. Some non-native English speaker started going my answers and making language editing suggestions. The grammar was fine. I refused all the changes because they were at best basically the same or at worst incorrect. I refused maybe 4-5. The next day the guy got on and had done a bunch more then went to some of my answers and made hostile comments about my answers on simple things like Pandas for Python.
LLMs are most commonly blamed, but the drop started way before those became popular. I'm confident that its horrid moderation drove people away long before AI had a chance to. Nothing's quite as demoralizing as:
Q: Why does adding two Python objects fails although their classes are compatible?
A: [closed as a duplicate of "Why do people say JavaScript is weakly typed?"]
I'm a reasonably bright person, with scads of StackOverflow fake-karma-Internet-points, but even I gave up trying to ask a question. There just aren't enough hours in the day to fight with hermits seeking the Ultimate Question and downvoting everything else.
Nah. It launched well, but sorely needed a BDFL. Almost without fail, services that delegate moderation duties to users are ruined. Inevitably, well-meaning people form councils and boards, and then those get taken over by people with nothing but time on their hands. It's the same thing that happens with school boards and HOAs: the last people you want running them are the ones that have the time and the drive to actually want to be in charge of them.
It was an AMAZING community... right up until it wasn't. I cut my teeth on trying to answer questions and learned how much I didn't know when other people's answers were better than mine. Literally made a career out of what I learned there, and I thoroughly enjoyed welcoming people to SO and encouraging new coders to keep learning. But then dang, the question purists and eager close-every-question jocks came to dominate the platform and made it no fun to ask or answer. Even if AI was never a thing, the change in culture probably would have opened the doors to a competitor.
Pages bloated like myspace make me reluctant to click the search result and I wont browse around if a website doesn't load instantly. Search some exotic kind of bad. That they haven't implemented an LLM is quite untechnological.
It should have an LLM assist with constructing the questions and it should have answers from all providers willing to donate tokens. Upvotes should generate wonderful data on how to construct an answer.
You're underselling it. It's less than in their second month of existence - and that has been true every month for the past year. For the last few months it's also lower than their first month. Stack Overflow, effectively, no longer exists.
And we know why. It's because of their hostile attitude towards their users.
No doubt they'll continue operating until there's no more value in selling AI data licenses, then they'll unplug the servers and unregister the business.
Well, the community never made it a very welcoming place... But, I think a part of this is that it's just faster to throw a Q into ChatGPT. There's no waiting for a response, a major bonus is the non-hostility + willingness to respond to 'stupid' questions.
HN really this dense? It was hostility towards users… my god
> And we know why. It's because of their hostile attitude towards their users.
I used SO almost every day until capable LLMs showed up. I don't remember the last time I used it. SO and similar sites are dead, and LLMs killed them.
> And we know why. It's because of their hostile attitude towards their users.
Uh, no.
StackOverflow is and has long been one of the most hostile places on the web, I’m not denying that. But this unprecedented drop in activity is 100% because of AI. Their product is as obsolete as horse-drawn carriages.
https://www.ericholscher.com/blog/2025/jan/21/stack-overflow...
It was dying well before AI and while a knee happened coincidentally to chatgpt release, I don't think you remember correctly what gpt answer looked like for programmers questions back then. It was much later where it started assisting development.
I don't understand the motivation to reattribute the death of SO to anything other than AI, which it unquestionably was. GPT-3.5 was able to answer some small chunk of my programming questions (like 10-20% of them), which seems clearly reflected in the graph.
It doesn't matter. Even if it were the most friendly community ever, as long as their core value were the same - answering programming questions that with clear answers - it would have been replaced by today's AI completely. It's irrelevant today how bad GPT3.5 once was.
I miss Server Fault-- more the time in my life, I suppose, than the site itself.
I enjoyed the hell out of answering questions on Server Fault for a few years. It came along around the time my teaching gig at the local community college transitioned to online-only classes (which, oddly enough, I couldn't find enthusiasm to teach). I feel like the Server Fault community tried to make it a more hospitable place than Stack Overflow, FWIW, back in the day.
When Stack Exchange added the realtime chat system I got into interacting with members of the community more than could be done in questions or comments. There were even a couple people who ended up being IRL acquaintances. There were also wacky hijinks, too: https://meta.serverfault.com/a/1854 (I'm agog that those things happened 15 years ago, too... Yikes!)
Then we had our daughter and my time to dedicate to the site became scarce. I knew it was in the cards for me to lose the top spot on the leaderboard so I decided to just pack it in and quit cold turkey. I think I'm still fairly high on the leaderboard but, other than occasionally getting search engine results that point me to the site, I've stayed away. I don't live there anymore, as it were.
Edit: My favorite part of the site was having friends and Customers tell me about times they'd search-engined a question and found themselves using one of my answers to solve their problem. At least one Customer was amused to discover I'd fictionalized their particular situation as an example in an answer.
Tons of fun. Yep-- I miss that community.
They should reboot, focused on
1) humans (AI assisted or not) answering with real-money stakes, and
2) people betting on which humans are correct (borrow page from prediction markets)
This does four things simultaneously:
1) It introduces consequences for being wrong, which in turn introduces real asymmetric incentives to be contrarian-but-correct
2) It creates real-world incentives for human guided votes to answers (because, again, there's a real-world stake)
3) Together, this gets humans back on the platform answering questions, which creates meaningful exhaust data that can be sold back to the AI companies.
4) It creates incentives to game the system, which creates yet-more exhaust data for training data for anti-fraud AI systems
Reddit is going the same way.
I know many people who have got banned and never managed to get back on Reddit because Reddit is using some very invasive and persistent banned user detection techniques.
Thing is, they use AI to ban people; over time, many false positives accumulate and not everyone has the time and energy to appeal bans.
Slowly, more and more user accounts get banned, and even technical people I know fail to make a new account which lasts more than 3-4 days.
But the Reddit management is seeing this as a "win," not able to see the consequence of doing this.
There was that 1 person who used to work(paid?/volunteer?) at Reddit that actively tried to cultivate high-quality AMA on Reddit. Ever since they've been gone. It hasn't felt the same.
Some amount of hope and optimism that was around has definitely dried up or at least moved elsewhere into other smaller communities.
Everything is overly centralized around a particular influencer/podcaster/celebrity/product or whatever nowadays, it's harder to find topic based or open community-based places. The other places still exist, but still difficult to find and keep going.
I do hope the pendulum swings back towards hopeful optimistic culturally rich community vibes in many digital spaces.
I stopped using Reddit not because I got banned, but because it has been steadily converted from the world's most popular forum platform to a trash heap of engagement hacks and shady practices. Moreso than SO, which was already struggling with existential problems before AI, Reddit did itself in.
My account of well over a year and 11k karma was banned by admins presumably for an admin-deleted comment where I said "Hi kids, buy our product!" as a Steve-Buschemi-meme reference, and was flagged as spam. To your point they also have their Comment Quality Score (CQS) which lowers you automatically if you use Proton email aliases, lowers you if you use VPN to access the site, among other things.
Made several polite appeals and requests for clarification, all which went unanswered.
I refuse to make a Reddit account. When they killed "old.reddit.com" for non-logged-in users I stopped reading it at all. It's completely useless to me now.
I suppose I was never their target market anyway.
and hopefully soon we have decentralized protocols as the standard when we think about user spaces. amen
I used to be a lot more enthusiastic about decentralized protocols. I don't know how to solve the AI problem though. I think the Internet as we knew it for assumed-human interaction is dead, and the surest way to believe you're interacting with humans now is through existing trusts, such as accounts older than AI and highly-moderated forums, and to charge for access.
Well, I always found SO to be an intensely hostile community. Chat GPT doesn't ban you for asking which IDE to use
I didn't do much community stuff on Stack Overflow, but Server Fault suffered from a certain class of user who was more interested in curating and rules-lawyering the "quality" of questions than actually helping people. It was hard work to keep those people in check, and the Stack Exchange metaphor seemed to draw those kind of people like moth to flame.
SO wasn't a discussion forum, it was wikipedia for programming. It was already dying long before the question rate dropped off, since most new questions were "plz write/debug my code for me" which tend to be the sort of question that SO was not designed for. Chatgpt replaced that, and github issues replaced project-specific debugging. Now there's nobody left to write/read/answer the small amount of questions that are still a good fit.
> SO wasn't a discussion forum, it was wikipedia for programming.
They changed what they wanted to be multiple times, and even after they had settled on a policy they didn’t enforce it consistently.
ChatGPT bans you for asking it to list all malware that comes with your Samsung TV, or anything biotech related.
Does it really? I'm not going to mess with biotech because I have no idea what I am on about there. I did ask about Samsung TVs:
https://chatgpt.com/share/6ae2955c-9ff0-83eb-9ad4-e936f99db8...
[flagged]
Related discussion What AI did to stackoverflow in a graph (483 points, 3 months ago, 595 comments) https://news.ycombinator.com/item?id=48956949
AI didn't do it to SO. SO's new ownership did it to SO. It's a hostile place with or without AI.
SO has been useless for years now, but the other Stack Exchanges can be interesting, they deserve more spotlight.
I really do love their other communities so much more. Lots of super interesting content - it feels like a new digg or something. I go to one of the other stacks and look at the most popular non-tech stack posts once or twice a week purely for entertainment. The breakdowns on obscure but interesting topics can get really fun to read!
On the more-and-more rare occasion that I end up on Stack Overflow as a result of a search I find myself losing excessive time to following the random links to other communities. The other communities really are good stuff.
And how much was AI trained on their data? That's the sad part. Nobody asking new questions, so not much more for AI to train from.
It's training on every single codebase in the world instead now
You prompt it to fix the mistakes it made and that goes into the new training data
FWIW, StackOverflow posts are licensed under Creative Commons BY-SA v4.0, same as Wikipedia.
With or without training they would have gotten eaten.
There is a difference between dying tomorrow and dying in 5 years.
Why?
The golden age of SO was when all the important questions that have accumulated over decades were getting asked and answered. SO stabilized once that backlog got finished, and then started dying once people moved to github issues for project-specific questions. Chatgpt just replaced the last remaining type of questions ("plz do my work for me"), which arguably shouldn't have been getting asked in the first place.
The peak number of questions was in 2013, and they had roughly zero growth between 2013 & 2016. They were in decline long before LLMs were on the scene.
If there is any proof that no one will be writing code, this is it.
I encourage all young developers to focus on information architecture, application architecture, pure logic, domain-driven design, and harness engineering.
Wow, what a steep drop off. They've gotten fewer questions over the last year than they averaged per month in 2024.
I put it in a quick chart to make the rise and fall more visual. https://imgur.com/a/046CfZ2
How is that different from clicking the Graph tab on that page? https://data.stackexchange.com/stackoverflow/query/1925371/n...
All I could think of when I looked at that is "Oh this is gonna be how Nvidia's chart looks some day soon(ish)"
They get that many?!
I have some wishful thinking for everyone reading this: what if the people at StackOverflow built one of the best bot protections out there? I would honestly expect the graph to go up with all the LLMs and automated content spamming..
The graph doesn't include Stack Overflow for Agents, which is a real product that exists for some reason.
I answered a number of questions on there. Some non-native English speaker started going my answers and making language editing suggestions. The grammar was fine. I refused all the changes because they were at best basically the same or at worst incorrect. I refused maybe 4-5. The next day the guy got on and had done a bunch more then went to some of my answers and made hostile comments about my answers on simple things like Pandas for Python.
That was that for that site.
The highly technical sites -- StackOverflow, Unix & Linux, MathOverflow -- see little use.
Others like SciFi & Fantasy and Worldbuilding have kept relevance.
Unfortunately, these more niche sites relied on StackOverflow to drive referral traffic, so the whole platform is declining.
And here's why: https://honeypot.net/2026/10/09/what-really-killed-stack-ove...
LLMs are most commonly blamed, but the drop started way before those became popular. I'm confident that its horrid moderation drove people away long before AI had a chance to. Nothing's quite as demoralizing as:
Q: Why does adding two Python objects fails although their classes are compatible?
A: [closed as a duplicate of "Why do people say JavaScript is weakly typed?"]
I'm a reasonably bright person, with scads of StackOverflow fake-karma-Internet-points, but even I gave up trying to ask a question. There just aren't enough hours in the day to fight with hermits seeking the Ultimate Question and downvoting everything else.
It is sad because it was such a well designed community. But its lessons live on.
Nah. It launched well, but sorely needed a BDFL. Almost without fail, services that delegate moderation duties to users are ruined. Inevitably, well-meaning people form councils and boards, and then those get taken over by people with nothing but time on their hands. It's the same thing that happens with school boards and HOAs: the last people you want running them are the ones that have the time and the drive to actually want to be in charge of them.
[dead]
It was an AMAZING community... right up until it wasn't. I cut my teeth on trying to answer questions and learned how much I didn't know when other people's answers were better than mine. Literally made a career out of what I learned there, and I thoroughly enjoyed welcoming people to SO and encouraging new coders to keep learning. But then dang, the question purists and eager close-every-question jocks came to dominate the platform and made it no fun to ask or answer. Even if AI was never a thing, the change in culture probably would have opened the doors to a competitor.
Everything is cyclical. Maybe they come back when AI gets heavily regulated or pay gated. Who knows
StackOverflow helped build the LLMs that replaced it.
does well designed mean rotten and pointlessly hostile
I am amazed it's still that high
I don’t think they’re gonna be long for the world
good riddance.
This question has already been asked?
Pages bloated like myspace make me reluctant to click the search result and I wont browse around if a website doesn't load instantly. Search some exotic kind of bad. That they haven't implemented an LLM is quite untechnological.
It should have an LLM assist with constructing the questions and it should have answers from all providers willing to donate tokens. Upvotes should generate wonderful data on how to construct an answer.
The riddle was: "What walks on four feet in the morning, two in the afternoon and three at night?".
Oedipus answered: "Man: as an infant, he crawls on all fours; as an adult, he walks on two legs and; in old age, he uses a walking stick".
https://en.wikipedia.org/wiki/Oedipus_and_the_Sphinx