Comment by cowlevel
7 hours ago
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.
> And we know why. It's because of their hostile attitude towards their users.
That’s why it succeeded. It was not a forum, it was something new. Contribute a new question that was well stated and broadly applicable and get a response. You had to earn your answers.
It wasn’t flawless, but it was good and so good it fed all the AI agents with good materials.
It wasn’t a “help me!!!111!” site, it was an exchange.
I put lots of effort into narrowing my questions, eliminating obvious answers by demonstrating the work I put in, and editted all my questions like an email to my boss. I was almost always rewarded with good answers.
> That’s why it succeeded.
The moderation style in the latter years was very different than at the beginning.
It was always heavily moderated, but after some point it felt like there were more layers of moderation and rules and cultural expectations added every year.
Moderation went from being a background cleanup operation that quietly kept the site in line to being the primary thing I expected to see when following a StackOverflow link from Google. In the later years it felt like every other StackOverflow search result was going to be the question I had followed by a moderator shutting it down for some reason before anyone could answer.
The moderation wore me out and I wasn’t even posting questions, just trying to find answers!
I asked a question about 15 years ago. Lots of responses/engagement and a good answer. I just got an email a month or so back telling me that question got closed as a dupe. I checked...it was not even close to a dupe. Exhausting.
1 reply →
> It was always heavily moderated
No, it absolutely very much was not. Even granting that you mean actions of the community (what we call "curation") rather than the small and extremely strained moderation team, way too much stuff historically got through, while people yelled at the OP in the comments and argued over downvotes instead of taking the action that actually matters: voting or flagging to close the question. Worse, people would post answers that acknowledge the problem with the question and asking OP in the answer text to fix things.
I know because I was part of that mess back then, because I didn't know better. That, in turn, was mostly because the entire set of ground expectations was poorly communicated, and the community wasn't given the tools to improve that communication (e.g. couldn't edit the help pages, had their feedback about the discoverability of site documentation ignored, etc.). But it was also because Atwood and Spolsky hadn't refined their intuition about why beginners' bad questions are bad. (Hint: it is not because most of their problems are easy to solve, or "beneath" experts.)
> when following a StackOverflow link from Google
Google got worse at search over the years, so you ended up seeing low-quality versions of the question more often. And the site got way too big (there is no reason why there should be more than three times as many publicly visible questions on Stack Overflow, as publicly visible articles on Wikipedia), such that search results could get diluted with irrelevant low-quality questions that happened to have a lot of keyword overlap with what you wanted. And misleading questions didn't get cleaned up, so you would search for X and find a question where the OP hadn't tried to analyze the situation at all and ended up asking about X when the problem was actually Y, and got the question closed as a duplicate of the main Y "canonical". (I know because I used to lose considerable amounts of hair trying to find the X canonical that I knew was there but stubbornly refused to turn up in searches.)
If "moderation" had actually "always" been "heavy" then you wouldn't have had this problem because this stuff would have gotten hidden from public view, one way or another.
> a moderator shutting it down
Again, these actions are overwhelmingly not taken by moderators, and phrasing things this way reflects a refusal to interact with the community on its own terms and understand its policies. Those who hold an outsider mindset are bound to end up feeling like outsiders.
Yeah, it makes me curious about exactly what thinking was involved in the site going to heck like that. Part of it was questions and answers accumulating and so the question of duplicates thus apparently weighing on the moderators minds more. Part of it might have been an effort to prevent people gaming the system.
Actually, talking to ChatGpt, I found this article: “Stack Overflow Isn’t Very Welcoming. It’s Time for That to Change.” ( https://stackoverflow.blog/2018/04/26/stack-overflow-isnt-ve...) and that actually shows what the SO admins didn't get (imo obviously).
They thought the problem was people not being nice. But that actual problem was that SO was "gamified" and it's gamified interface made it impossible to win unless someone was ultra-competitive (and I'd those types burned-out over time as well). People not being nice was real but it was side-effect of the situation (questions answered instantly so a would-be question answered had to be constantly racing).
It's easy to imagine how the admins didn't see the situation. Each up of intensity seemed to up the engagement and so seemed a good thing. The question of whether the process was sustainable didn't appear and when there were problems it was attributed to something else.
I think
> I was almost always rewarded with good answers.
Huh. I had the opposite experience. The better my questions got, the fewer responses I got.
I stopped bothering, after a number of questions went answered only by me (after I found the solution on my own).
It did reduce the insults, but that seemed to be the only gain that I got, for carefully articulating my questions.
LLMs aren't perfect, but they don't imply that I'm being lazy, for not already having the answer, and they do give me an answer, instantly (and, like SO, the answer is not always the best one).
I loved it when I iterated on my question so much that I accidentally solved it before I got to posting.
I have engaged with some younger devs who don't even bother turning the thought into a question. They make statements and expect you to do the work of inferring the question. Which I feel like they got from chatbots since it's faster and LLMs don't care if you make them jump through hoops.
> The better my questions got, the fewer responses I got.
There is a sweet spot of providing just enough detail but not too much in asking a good question. And even the best only “bat 300”. I've posted answers to my own questions plenty of times when I got no response.
I hear you on the hostility. Also a lot of people didn't experience how bad things were BEFORE stack overflow. The first time I tried asking a question on IRC I got mocked so bad I logged out and never tried again. Most forums were like that.
I gravitated to Ruby back then because it was one of the only forums people would be patient and empathetic with me.
5 replies →
I found that Stack Overflow, like Reddit, rewired my perception of what "good" is and made me view bad things as good.
60000 rep btw
Was it a better question for skimming and quickly replying,
or just a great form of your question if someone genuinely wanted to answer it, not pump their score, dopamine etc?
It certainly didn't start that way, having been there and followed them on their podcast -- pre-launch. It became more and more hostile. It was initially positive, even when questions didn't fit the mold. Then successive waves of pedantic moderators chased out the altruism. There never seemed to be a suitable mechanism to slap down the haters, then years pass, you need to ask a question, and find out that it is causing you anxiety to just post a question..
I think the demise of the SE network (not just SO) owes a good bit to this rigid application of the "hacker" mindset (more or less) to a lot of fields that do not benefit from the hacker mindset.
From my experience in Home Improvement and Bicycles, I almost always get in some form of contention with the mods over the boundary between "broadly applicable" and "too specific". I'd do the esr-minimum[1] of showing my homework by posting pictures of my WIP and citing Youtube videos I'm trying to follow but, apparently, this is already toeing the line. I honestly don't know what the mods expect of a good question there. Of course, there will be an appreciable degree of case-specificity in my questions because if the YT tutorials could already answer my problem, I would have no need to ask for help and advice!
Then we have sites like Literature SE which has been languishing in their sandbox since literally forever. Like most SE sites, the guidelines bid you to "avoid asking for opinions or open-ended discussions" and "be specific" but I find that untenable in a Literature site. Notably, sister site SF&F is comparably thriving but there is a fair amount of open-ended speculation going on in there.
Lastly, IMO, it just suffers because at its core its problem is human and social, not technical. In an effort for better curation we ended up with Computer Science, Theoretical Computer Science, and Computational Science. For some posters, finding the right SE site already introduces a huge wall to overcome for posting. This hampers a lot of the out group: students who might be seeking homework guidance or looking to supplement their own readings and lectures, amateurs who are just curious about something in the field. I've had a question in one of the Science SEs outside my field that was well-received by the community but one mod kept "tagging to delete" (I forgot the exact term) almost every year. This mod action has a prescriptive period and, boy, he kept at it. Honestly, I don't care in the end. I asked the question out of sheer curiosity and got a satisfactory answer. Yet I wonder how exactly they expect to foster a community this way.
[1] http://www.catb.org/~esr/faqs/smart-questions.html
> That’s why it succeeded. It was not a forum, it was something new.
And _then_ it became so much more hostile to users - and lost a lot of what made it so wonderful. It became very common that asking a new, good question would be marked as a duplicate because there was something it wasn't a duplicate of.
My experience with the late SO was, that the original question used to be ten years old and filled with responses that were no longer relevant to the current state.
What amazes me is that an LLM can actually make sense out of all this crappy training data and often produces meaningful answers. (maybe that's a sign that we already have AGI)
You'd also get people downvoting completely valid rule abiding questions, and the site heavily punished being downvoted. So you'd get hit with a stick repeatedly for trying to engage with the site, it's no wonder everyone left as soon as there was an alternative.
At the start moderators worked with you, after 2014 they worked against you.
When Attwood and Spolsky founded it, they commented on it being because Experts Exchange was hostile to users. https://news.ycombinator.com/item?id=3656581
If StackOverflow is being considered hostile now it's big ... well, Google the three stages of enshittification I suppose.
Ah I miss the old days of expert sex change.com
To be fair not really but it’s a name I haven’t heard in a long time and all it took was to scroll down to the end of the page to get your answer.
Real 20 year ago internet there.
Agreed, being a n00b in the open source & hacker community I recall asking half assed question on forums and IRC and would get the infamous RTFM.
I detested it but quickly realized that I needed to put in work and learn and only when I understand can I ask valid questions to real problems…not just asking for someone to spoon feed me an answer.
I really appreciate that experience…
> I was almost always rewarded with good answers.
Really? Because I only ever asked three things, technically two I suppose, and the only significant interaction I got from there was a guy saying they thought the third question was a duplicate but they only didn't mark it that way because they saw it was me asking again.
This dude straight up admitted to be on the hunt for marking things as duplicate and only put brainpower into doing not-that because something felt off in is gut.
Then someone else did.
No, he didn't actually answer my question. No one did.
> And we know why. It's because of their hostile attitude towards their users.
Right, their decline is because of the thing that they've been accused of since the very beginning, and not the trillion-dollar technology that replaces them. Brilliant take.
Let's just be blunt. The main target audience were clueless idiots who just wanted to solve their current programming problem, e.g. compiler error etc. They didn't care about anything else, they just wanted to copy-paste text and for the problem to go away, preferably without even needing to turn their brain on.
Now we have a new guy in town, who offers "robotized copy-paste the solution to YOUR specific programming problem, no reading required! we can even control your computer for you, so you can do something you actually enjoy and stop wasting time on this complicated computery thingy for boring nerds", which serves the aforementioned audience much better.
> The main target audience were…
"Target audience" connotes people that the site wanted to be there. That's just who showed up, because that's who will always show up basically anywhere that looks like it might be helpful to them, by default.
1 reply →
are you perhaps a mod from SO? if not you should apply as one asap.
you sound exactly like the mfs that ruined that forum.
3 replies →
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.
That and [redacted] mods who delete your question “because its a duplicate” but its not, you just assume it is and lock it. Lots of this when .NET Core came out and it drove me up a wall. Wouldnt surprise me if a lot of “hallucination” type answers come from bad Stack Overflow data being interpreted as “fact” in earlier models.
Your question was neither deleted nor locked. It was closed as a duplicate, which immediately provides a link to existing useful information and helps keep the site organized by putting the best answers in one place.
My extensive experience has been that the overwhelming majority of the time that people complain "this isn't a duplicate", it actually quite obviously is. (And there were even more irritating complaints on the meta site at times; all the way up to the level of "you shouldn't ever close duplicates at all because I think you shouldn't", demonstrating no interest in even trying to understand the reasons for doing so.) Of course, people do make mistakes; there was an entire process for fixing this, but most people could not be bothered to engage with it productively (or even civilly). But feel free to give examples of things that you feel were wrongly closed from your own experience.
It's astonishing how vastly different our experiences were. I trawled the new Python questions frequently, looking for fun, novel things to answer. It was dismaying how often they'd been wrongly marked as dupes, often because the original answer was ancient and obsolete. It would often be egregiously bad, like:
Q: How do Python's dict preserve insertion order?
A: [duplicate of "Why does a Python dict's keys have a different ordering each run?"]
It was clear that the mod/curator/whatever spent no time attempting to understand what was actually being asked, and linked it to something that was completely incorrect as an answer to the real question.
I saw this constantly, so frequently that I'd be hard pressed to answer with specifics. If you asked me "when did you last eat a fruit bat", I can tell you the month and year because that stood out. If you asked me "when did you last eat spaghetti", I'd have a hard time saying.
Not just their users. My org got pitched stackoverflow careers and their private knowledgebase around 2019. We were somewhat interested but found it too hard to work with them.
We weren't going to be a huge contract (~100k/yr) but the tone was very much "conform or fuck off". I think the sticking point was that we were offered 1x permanent careers ad in North America and 2 in Oceana, and we wanted 3 in Oceana instead.
I've had a similar experience the couple of times I've asked or attempted to respond to questions (about software I wrote lol) - grammar and framing seemed more important than technical information/a solution.
N=1 anecdata ofc, but I'm not surprised at their demise
> 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.
They were on their downswing but LLMs put a gun to their head and executed them. I don't use Stack Overflow anymore, just an LLM and cppreference.
Damn sad how quickly that habit changed for me.
Never asked a question, but got some good info from people who had asked questions and the people who replied.
Yeah, it has nothing to do with 'hostile attitude'. There's just no reason to use StackOverflow when I can use a chatbot that gives me infinite StackOverflow answers for any conceivable problem I have, within seconds.
> It's because of their hostile attitude towards their users.
The same thing that happened to perl.
Perl was my first love as a language, and I don't recall any hostility towards Perl users. On the contrary, the environment in the early 2000s was very supportive.
Perl likely declined because Perl6 did a Duke Nukem Forever, and Perl5 was dated and not incredibly well-adapted to OOP. Then Python got taken up for teaching, and it was all over.
(I still love Perl5, still use it in one-liners all the time, and still code up quick CLI tools with it.)
How so? I stopped hanging around in Perl-land around 2001 and it seemed friendly enough way back then. What happened to it?
I remember seeing at least one person complain that perl forums were unfriendly, but in that case it turned out it was because he took it as an insult when after he posted code others would comment with re-writes using fewer lines/characters. He didn't understand that they were just playing and not critiquing or trying to tell him that his code should be have been written that way instead. If anything, the shortest possible way to do something should probably be avoided in perl code you plan on ever having to work on again, but it can still be fun to see how compact you can make it.
1 reply →
> 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.
People used to add "reddit" at the end of their search queries, and now I'm sure many of those people just ask a bot. But reddit is still there (even if in the form of a grotesque mutation of itself) because people like to talk to each other.
So yeah, AI surely didn't help, and I concede that the SO case is even worse given its nature (and how good of a substitute AI is), but if it wasn't a pool of acid I could see it existing as a place for people who want human help and opinions. So I don't think it's fair to say the drop is "100% because of AI", as you put it.
Then again, it's been a pool of acid since forever and it kept going for a long time like that, so maybe you are both right. AI did kill it, but because it was hostile.
> But reddit is still there (even if in the form of a grotesque mutation of itself) because people like to talk to each other.
I hear they're drowning in AI bots. I guess a fair amount of people also don't mind talking to those.
But SO was never intended to be a place where people "talk to each other". It was specifically intended and designed (sort of; the design was faulty) to avoid chatter because that gets in the way of others coming by later and finding a question and reading its answer.
To me your description is mythical reddit. Actual Reddit is every bit as hostile as stack overflow and unlike stack overflow, old questions can not be updated with new answers when existing answers get out of date.
8 replies →
But SO was never a place to just talk to others. It's not a general purpose discussion site. It could have been a friendly place, but users still would have left, because it's just far easier to get an immediate answer from AI, and those answers have only been getting better.
Yeah, the place could be very hostile, but that hostility didn't crater questions before ChatGPT was released.
2 replies →
SO responses were peer-reviewed code snippets or fixes. If something had 800 upvotes and comments outlining pros and cons, that is useful information. The bot gives you whatever it wants to give you, it could be information from a scientific paper or late night TV infomercial.
SO's new-question rate peaked in 2014 and has been declining since 2017 (with a small, short-lived, early-COVID bump). LLMs have certainly hastened its decline, but it would still be in a pretty sorry state today even without them.
the obvious reasoning is that the vast majority of common questions were already asked, and then still many, many uncommon questions, until the only things that hadn't been asked were very, very niche. It's the same reason complaints about mods marking lots of questions as duplicates didn't really pick up momentum until the mid/late 2010's.
People like me⁰, who I admit will also soon be as obsolete as horse-drawn carriages, will still refer to it for some time.
--------
[0] I'm leaving tech as a career in the next year¹, I don't care if I have to wait tables on minimum wage, I've been unhappy since ~2020 for a growing number of reasons and the great purloining+cheating+polluting machines are the final bail of hay
[1] Maybe before then end of this year, my notice period is two months and I have some accrued annual leave to use, we'll see how the next week or so goes!
Good luck. What are you thinking of moving on to?
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.
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 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.
10 replies →
Because AI removed what little value it had.
AI isn’t magic and can’t always give you the right answer. And it can’t give you the why a lot of the time.
People’s actual experience is useful, and stack overflow was great for that. Until they ruined it.
If the site hadn’t been ruined, it still would’ve had value today. Traffic would’ve dropped, but not to near zero.
AI can’t always give you the right answer. But it tries. StackOverflow stopped trying to answer questions.
It's like how everyone hated dialup but still kept using it anyway until there was broadband.
Their product is as obsolete as horse-drawn carriages.
The product isn't obsolete. It's just been re-packaged by AI companies in a far more useful form.
It is not obsolete insofar as it is a good source of Q/A material, which is useful to both humans (directly or indirectly) and bots. The only part of it that is obsolete is the search/query part. If the AI companies had an ounce of decency, they would see no problem injecting money into the platform to continue seeding that Q/A material which they have directly profited from. But investing in a commons is not on the playbook for vulture capitalism.
[dead]
[flagged]
[flagged]
So you're arguing that SO was really un-hostile at first but at some point it became hostile, causing the crash in usage? My recollection is that SO always was very hostile towards users, even during the time period when it was growing rapidly. So hostility alone is not a sufficient explanation.
Other question asking platforms like Reddit and Discord are still going well. Nothing has been hit as hard as Stack Overflow, because people hated having to use the site the whole time.
sort of like the hostility you're showing?
Punters with enough good boy points were rolling into very obscure and technical tags to close questions as duplicate (they weren't). When I saw how impossible it was for people to ask questions in a tag I was an expert in I knew the site was cooked. Imagine, like, giving physiotherapy forum power users moderator privileges on HN.