Same here. LLMs are great at spitting out well-known solutions to problems instead of the best one. The "long tail" of solutions is usually lost due to how tokens are sampled from the LLM's probability distribution.
What I found to help a lot is to ask for e.g. 10 different solutions to a problem and then choosing one of them. Sometimes, this even leads to borderline creative solutions if there aren't 10 different ones.
In theory reasoning tokens should do the equivalent of this - explicitly create options outside of the quick-response probability space, so those can guide future generation.
In practice, models that do this won't be prioritized as much, because the economics of thinking tokens that stop by default at, say, one option plus a bit more planning (short of full alternatives) would be superior as long as billing is per-user instead of per-token. So we'll still need to play games with prompting!
> LLMs are great at spitting out well-known solutions to problems instead of the best one.
I remember how Stack Overflow would close questions as duplicates just because somebody suggested the wrong answer that is also the right answer to the existing question. The best way to get a correct answer on Stack Overflow (and forums before that) was to post the wrong answer as part of your question.
One thing that SO had was you could see multiple solutions and implementations for something. Sometimes the "best" solution isn't very readable code, sometimes you are able to understand the problem better when you see a bunch of people solving it in different ways and arguing about it like angry monkeys.
SO has always had a pretty strong stance against opinion-based questions, but this is maybe the niche they should be exploring now. Humans still have a lot to say about the "best" solution to a given problem. The whole idea of an "accepted" answer could be removed, for example, since that's what AI will already generate.
needed to implement a language feature that was a bit complicated and im not familiar with it so just planned with claude to do it, and after each write/fix cycle it just wouldn't work right.... gave up, went back to SO copy pasted the (not perfect but enough to start from) answer and worked up from there...
at the same time my knowledge grew and im more confident to do this same capability myself whereas reiterating with claude it was just a slog and i didn't learn much...
i think i may be starting to sour on these "do it all for me" usage scenarios for ai... especially for unfamiliar areas...
Claude does it quite a bit when you’re triggering the search tool functions.
It’s fine, and what you would expect for certain prompts, except that the synthesized results often come back communicating more authority than they deserve.
It was funny for me, when I asked it about something specific exotic - and it gave me a confident answer. But checking the sources I discovered it was from my own inquiries on a forum thread about it from the last time I unsuccesfully tried this (before the agents came) And so I knew, that any authorative tone was undeserved.
On the other hand, Claude later nailed this project, where I as a human said before, no, too much extra work.
I had email correspondence once with a vendor about how to talk to their i2c bus. The documentation was all asm, and I wanted to at least “uplift” to C. They didn’t have any answers, so I sent them my solution which was was the asm calls that the c stdlib decompiled into.
4 years later my company had bought a different company, who happened to be using a newer model of the same board. They asked me how we could use the 12c bus. “Well before you bought us, we emailed the vendor and sent back this C snippet”
It was my code, verbatim. I’ve always wondered how many times they passed that bit of code around.
What you are noticing in a long term is the "community" knowledge and communication which the chatGPT is now kind of destroying. In some sense, it is no different from the difference between studying along and studying with your peers at a university.
You can definitely study alone and achieve perfect grades, but studying with your peers is how you build relationships for future life and take your community forward as a whole.
I am thinking to make canned encyclopaedia of stackoverflow answers.
Claude/Grok/Gemini/Chatgpt answers are often so… how to say it… misleading? I have to stop the conversation as it leads nowhere (and it is not a skill issue :)
well, SO is probably the highest quality data source for a language model and the rest of the internet is just diluting the final latent space limited by Jon Skeet.
https://software.codidact.com/ was created after one of the many SO dramas. It doesn't come up in searches though and I didn't have reason to use it...
Definitely not better off. SO was fairly mean spirited, but nowhere else has such a vast trove of high quality answers to common software problems been collected. SO likely trained many of these models with its answers, and I don't know what software development will look like when it dies.
No where else has such a vast trove of high quality answers been hidden because the question was closed as duplicate when someone else asked that question later.
Thanks for this post. Unfortunately, you used the wrong word choice here and this question has 13 other answers that have some of the same words but don't really answer your particular question so it has been deleted. Also, if this remains posted, my not-on-point answer will get less views.
Everyone loves to say this when the death of Stack Overflow is discussed, but it always was that way. Strict moderation, love it or hate it, was part of the platform. And it could have kept going that way for many more years if not for LLMs 99.9% obviating the need for a coding Q&A forum.
Everyone loves to say this... because it's everyone's experience. I stopped using SO as a resource years ago (well before the advent of LLMs) because it got to the point where almost invariably, when I found a post that managed to perfectly articulate my question, it was closed as a duplicate of some other, distinctly unhelpful question. But it wasn't always that way. There's a fine line between strict moderation and draconian moderation, and at some point they crossed from the former to the latter.
Don't forget the other way of sidetracking what you're asking for: "Why are you doing this, do something else instead."
I think most of my questions ended up with this, when I had very good reasons for doing it the way I was doing it. I typically wasn't showing it because I had isolated the problem I was facing into the minimal amount of code to duplicate it, or I was stuck with the particular tech I was using and we had 12 years of code built on top of it and I couldn't switch.
They also neglected to solve the problem of differentiating answers to previous generations of software. How many python2 answers does one have to sift through to get a python3 answer - maybe the weight of answers finally tilted over the probabilities. Even just adding the right tags would have made it easier, but it wasn't ever solved in any way as far as I could tell. And the old answers are there like potholes to fall into.
I rarely posted questions on SO but I largely stopped using it as a resource because of exactly this. I got tired of searching for answers only to find closed-as-dupe questions.
I feel like in their search for “quality” they completely forgot that they needed engagement to deliver value. The whole premise was that the correct answers would bubble to the top, but their system ended up pushing everyone to old questions that had a highly upvoted but either out of date or not applicable answers.
What’s crazy is with the current gen of LLMs you could 100% rein in the abusive mods. SO could reinvent itself as a place for coding agents to get clankermotiomal support after they ‘hear’ ‘WHY THE F DID YOU DO THAT’. (Seriously a hook could draft a submission and then set up a listener for responses. Who would be answering is an interesting question tho)
Ah yes, programming: the discipline which famously only has singular right answers to problems, such that programmers never get in arguments with each other about the correct approach to solving a given problem, and there are no long running disputes that have ossified into intractable disagreements.
Ironically i'm probably a better dev purely because after a few experiences on SO, I would rather waste days/weeks banging my head against problems and learning from them than to actually post on SO. It was a miserable experience generally. For context this was probably ~15 years ago now.
This isn't necessarily to say that SO made me a better developer. Rather i'm just saying that i value (correctly or not) those extremely hard fought lessons. Those lessons where it was considerable pain, effort, time, misery, etc. Are they efficient ways to learn? I doubt it. But in my many trips down that road i developed intuition that i'd probably not have otherwise.
So ironically i guess SO made me a better developer by avoiding using SO at all cost. Conversely, i imagine i'd lack this value that i speak of entirely if i was 20 years younger and starting fresh today. Not sure i'd be better off though.
edit: By "using SO" i should be saying posting on SO. I of course searched and used data found on SO as often as i could. So to that end i am grateful for SO existing.
Sure, the mods were not always the best on SO. But even if you did ask a question, you had to deal with a userbase that was more pedantic and judgy than Reddit. Usually you would get an answer if it was obvious, other times you would have to defend your question against some guy whose newfound obsession was whether you had an XY Problem. Or who was personally offended you weren't using whatever the fad library of the day was (e.g. jQuery).
> Against some volunteer who's encountering their fourteenth clear XY problem of the day.
Fourteenth clear as imagined in their head XY problem of the day.
By far most of the "XY problems" I saw, on SO or elsewhere, were actually "XY problem problems" - i.e. a responder having so limited imagination and character (or, to be charitable, just running very low on energy and focus), that upon coming across a question they couldn't comprehend, they would assume the person asking the question must be confused instead.
If it isn't fun to do, and simply causes frustration, that hypothetical person constructed in the comment could just step away for the day.
I get that dealing with low quality questions wasn't great, but imagine spending an afternoon researching a weird thing using some tools your organization mandates, writing it up, only for that person to skim it and just assume you really wanted to do $otherThing.
This. And it's even starting to be a problem with LLMs - noticed that with Claude and Gemini this week.
Yes, I am specifically asking if it's possible to do X with Y. No, I'm not interested in how to do ${unrelated except for name} thing A with Y, or ${manual variant of X} by hand to ${subset of Y}, nor do I want to use tool Q instead. I specifically want to know how to do X with Y, for reasons that are my own and borne of frustration with Y being a toy I'm trying to use for productive work, which apparently means pushing it past its operational envelope, but I have a deadline...
X in an XY problem is almost always possible, like "how do we make ship canals and harbors with nukes" (https://en.wikipedia.org/wiki/Project_Plowshare) or "I'd like to amputate my hand to fix a hangnail".
I don't think that's true. I remember the very early days of Stack Overflow and it felt much more fun and friendly than it did 6-7 years later. I have so many 15+ years question/answer that somehow get revisited by a "moderator" that decides that maybe we should close this.
But was that the cause of Stack Overflow's demise? I agree that it most likely isn't. It's most definitely because of LLMs.
A sink that has large but finite capacity to absorb something can reach an irreversible tipping point when an additional shock happens.
There are many examples of this in nature. (And in Nature [1].) One interesting one that I think is unknown to many people is limnic eruption. A lake can absorb quite a lot of CO₂, for example from volcanic gases. Dissolved CO₂ is invisible, so the lake can look quite ordinary, but the build-up turns the lake into something approximating an unopened carbonated soft drink. If the lake is deep enough and the layers don't mix frequently enough to relieve the pressure, it can build up to the tipping point where the lake will suddenly explode, flooding the nearby landscape and releasing an invisible CO₂ cloud, which will proceed to kill the surrounding life by asphyxiation.
The conditions required for a limnic eruption are rare, though there were two incidents in Cameroon in the 20th century.
It's entirely possible that the build-up of hostility on Stack Overflow were survivable as long as it didn't build up to a level that exceeded the community's ability to absorb it. But an exogenous shock or the community shrinking could upset the balance, with hysteresis making the change difficult to reverse.
if early stackoverflow moderation was strict it was strict in a way that was invisible to people who genuinely needed the help. later on it got people who thought the strictness was the main point, and that they had to be vigilant defenders of the purity of the site (wikipedia had a similar malaise).
I personally gave up on the site entirely when I saw a very valid question from an inexperienced programmer closed as a duplicate and redirected to a question about a similar problem that did not actually address what they were asking.
I'm all for a heavily moderated forum (that's why I like HN) but SO was clearly on the decline (from the graphs in this post) before LLMs came onto the scene. It peaked around 2017 (not counting COVID numbers) and was in a steady decline. ChatGPT just pushed it off the cliff (figuratively and literally).
It was not always part of the platform. Moderation got increasingly aggressive on SO as time went on. There was an inflection point around 2014 where lots of people gave up on using the site. SO was in slow decline for most of a decade before AI showed up to finish the job.
Not to go off topic, but there's some similarities between that and the way that Hacker News is run/moderated. But I believe they've found a pretty ideal balance. Even though they occasionally annoy me with with fun things taken down, I understand the need to have some consistency.
Perhaps they need to take a page out of dang (and team)'s book.
Yeah sure but in the past there was no viable alternative so people tolerated the crazy moderation. As soon as AI offered an alternative people left in droves.
It's definitely plausible that if it hadn't been such a hostile place to ask questions (sorry ItS nOt a Q&a SiTe) that it would have survived AI better.
Everyone loves to say it because it's true. Asking questions on SO was always at best an adversarial process, but it got really bad in the last decade or so.
It was quite simply a profoundly unpleasant "community" to interact with.
Yeah, as I understand it, they wanted to optimize for Google search. This meant having "canonical" answers. This killed the site in the long term. In the short term, it worked wonderfully, and the founders made a (well deserved) killing.
Agree. But I asked a couple questions about a year ago and got zero response. It's not just people asking questions who deserted SO, it is also people answering them (probably a chicken and egg problem - reversal of network effects).
if the mismoderation did not kill stackoverflow, it at the least made people who might otherwise have supported the site feel like nothing of value was lost.
The author labels COVID and the launch of ChatGPT on the graph, but fails to mention that Stack Overflow was acquired in June 2021 by Prosus, a Dutch private equity firm. That looks to me like it matches pretty well with the entire downward trend.
It’s not necessarily the sale. Some private equity companies move from “Let’s invest like we’re shooting for the moon” to “Let’s invest like we want to improve margins and flip this on 3-5 years”
It’s not inherently wrong but it is a different model, and sometimes companies suffer as a result.
Businesses (and any other kind of asset) are sold for all kinds of reasons, and trying to time the market to maximize the price is only one of them. Probably not even the most common one.
Stack Overflow with all of its shortcomings was a marvel of the internet at it's peak. People especially in early were chasing karma and anything you asked, you were sure to get some answer. Not always right but some answer. While for sure LLMs will give much better answers on average. I feel that it's a piece of humanity we've lost there that should be adequately remembered and the memory cherished.
Wouldn't this be worrisome? People used StackOverflow and generated new knowledge along the way. Without such medium for discussion, how can we feed models with up-to-date quality knowledge?
How do you convince people to not want an instant answer? Even if SO didn’t result in so many “What have you tried?” responses and immediate closures, most people would still prefer instant feedback.
I wrote multiple answers to questions that weren't just "point to docs". And even when it is pointing to docs you are providing the reasoning as to why it works one way or another.
Ever since the AI stuff started rolling around on coding i've seen MORE documentation, theres a big incentive to properly document your API endpoints so LLMs can figure it out from specs, and even when not documented the llms can also just read the code and figure it out directly (for libraries and similar). And at least in my experience they tend to document or write it down for future sessions too!
on the contrary, theres more of an incentive for apis to have docs for agent discovery. the docs / interfaces themselves can be auto-gened (stainless / mintlify)
LLMs would post solutions to the issues that they've discovered after doing a lot of research.
Unfortunately the LLMs are concentrated into few providers (OpenAI, Anthropic, Google) so there's a chance they each end up doing their own private (and closed) StackOverflows.
By leveraging their private StackOverflows, their LLMs will be able to short-circuit complex reasoning, saving tokens, time, and money.
I knew that stack overflow must be suffering because of AI, but I find it hard to believe that questions asked per month has gone from 200k pre-chatbot to (what appears to be) ~1k. Although, I suppose I have not gone there at all in the last 4 years...
Clicking through to the query for the first chart, I see the peak of ~300k in May of 2020, and it was ~3k in April of 2026 (the last complete month). I’m flabbergasted.
Stack Overflow might be the greatest receptacles of human knowledge on programming.
But I would argue that it usefulness only extends to its body of knowledge. As a service and/or community it has been pretty terrible for a long time:
If you were a new user trying to learn programming, it was maybe one of the most toxic resources available. I don't think I have posted a question since 2019. And even there, the only thing the average user could expect was a snippy response from someone who barely stopped to read your post. And/or a mod deletion because a similar-ish question already existed (regardless of whether it had a satisfying answer).
At a certain point, all the meaningful questions have already been asked. The site exists to collect novel new problems and not help people with iterations on existing problems.
(Also, underrated is the extent that the industry has homogenized around a couple of frameworks that are used for everything. I think it's telling that the peak of StackOverflow coincided with the era that React was taking off, to just name one).
Early years SO was optimized for people helping people. Later on they ruined the site by optimizing for tidiness ... and griefing users (especially new ones) off the site in the process.
LLMs are better than slow human support of any kind for debugging / helpdesk work (which was never that welcome at SO anyway).
Stack Overflow is still great for canonical questions, multiple answers, public / SEO'd discussion between humans, etc.
But that probably isn't enough to save the company as a private equity acquisition hoping to 100x their $1.8 billion investment.
Hopefully the classic Q/A site eventually gets written off and spins into a Wikimedia-like foundation that is interested in preserving the original Q/A site and has no desire to grow or become something else.
There are prowling bots trying to strike up engagement with stupid open ended questions "do you find that using a golf simulator improved your golf?"
And some subs seem infested with submarine advertising, posts that mention a single product name almost in passing.
Nearly always these people have their posts hidden. Reddit has always been looser, people can edit and delete their comments and entire posts, and enjoy some frothy conversation while hiding their old rants.
There are plenty of signals that reddit could use to push out bots but they just don't seem to prioritize it.
When you find your self wasting time responding to a bot it's a bit of a sucker punch. Too many of them and Reddit will be on the ropes as a wasteland.
Just look at the graph, Covid peak aside its previous peak was in 2016 and it was in continuous decline since then. All LLMS this was increase the slope.
I know everyone here is heckling calling it a forum, but this is a basic forum feedback loop -- the forum activity declines, so people show up to check less often and get fewer responses to their posts, so the post there less often, repeat until traffic falls near zero.
You may have like a handful of weirdos who never leave and develop their own little community in the wreckage, especially if the cost to continuing to run the forum is trivial, but it's basically a death spiral every time.
one of the interesting properties of public forums like SO is that the maintainers of software packages get visibility into the problems and frustrations that their users experience
we're losing that signal when the Q&A behavior shifts into language models
i haven't been on stackoverflow for maybe 6 months, haven't seen it on google search, or needed to ask much on google either. Maybe they did a web search. On the plus side, its still used by AI agents
Call me crazy but sometimes I still find a better solution on StackOverflow than what Claude Code insists to do.
I'm not sure we're better off without SO in the long run.
Same here. LLMs are great at spitting out well-known solutions to problems instead of the best one. The "long tail" of solutions is usually lost due to how tokens are sampled from the LLM's probability distribution.
What I found to help a lot is to ask for e.g. 10 different solutions to a problem and then choosing one of them. Sometimes, this even leads to borderline creative solutions if there aren't 10 different ones.
In theory reasoning tokens should do the equivalent of this - explicitly create options outside of the quick-response probability space, so those can guide future generation.
In practice, models that do this won't be prioritized as much, because the economics of thinking tokens that stop by default at, say, one option plus a bit more planning (short of full alternatives) would be superior as long as billing is per-user instead of per-token. So we'll still need to play games with prompting!
Without continuous feedback from real world, lower-probability token (and soon high-probability ones as well) will be complete garbage.
> LLMs are great at spitting out well-known solutions to problems instead of the best one.
I remember how Stack Overflow would close questions as duplicates just because somebody suggested the wrong answer that is also the right answer to the existing question. The best way to get a correct answer on Stack Overflow (and forums before that) was to post the wrong answer as part of your question.
One thing that SO had was you could see multiple solutions and implementations for something. Sometimes the "best" solution isn't very readable code, sometimes you are able to understand the problem better when you see a bunch of people solving it in different ways and arguing about it like angry monkeys.
It really could be bad though.
SO has always had a pretty strong stance against opinion-based questions, but this is maybe the niche they should be exploring now. Humans still have a lot to say about the "best" solution to a given problem. The whole idea of an "accepted" answer could be removed, for example, since that's what AI will already generate.
The bot can do that kind of thing, too.
"Show me 6 very different solutions, and present arguments for/against each one as if a bunch of angry monkeys."
was just gonna post the same thing
needed to implement a language feature that was a bit complicated and im not familiar with it so just planned with claude to do it, and after each write/fix cycle it just wouldn't work right.... gave up, went back to SO copy pasted the (not perfect but enough to start from) answer and worked up from there...
at the same time my knowledge grew and im more confident to do this same capability myself whereas reiterating with claude it was just a slog and i didn't learn much...
i think i may be starting to sour on these "do it all for me" usage scenarios for ai... especially for unfamiliar areas...
Much of what Claude insists you do probably came from SO.
Or Reddit. I don't know about Claude but Gemini has given me answers that are verbatim comments from Reddit.
Claude does it quite a bit when you’re triggering the search tool functions.
It’s fine, and what you would expect for certain prompts, except that the synthesized results often come back communicating more authority than they deserve.
It was funny for me, when I asked it about something specific exotic - and it gave me a confident answer. But checking the sources I discovered it was from my own inquiries on a forum thread about it from the last time I unsuccesfully tried this (before the agents came) And so I knew, that any authorative tone was undeserved.
On the other hand, Claude later nailed this project, where I as a human said before, no, too much extra work.
I've gotten my own answers given back to me for problems I forgot I already had.
I had email correspondence once with a vendor about how to talk to their i2c bus. The documentation was all asm, and I wanted to at least “uplift” to C. They didn’t have any answers, so I sent them my solution which was was the asm calls that the c stdlib decompiled into.
4 years later my company had bought a different company, who happened to be using a newer model of the same board. They asked me how we could use the 12c bus. “Well before you bought us, we emailed the vendor and sent back this C snippet”
It was my code, verbatim. I’ve always wondered how many times they passed that bit of code around.
I've gotten this too a lot. If you ask AI to cite where it got info you can lose a lot of confidence in it pretty quickly.
I have seen it quote my own code back at me, including comments word for word.
I've seen chatgpt word for word plagiarise stack overflow answers
What you are noticing in a long term is the "community" knowledge and communication which the chatGPT is now kind of destroying. In some sense, it is no different from the difference between studying along and studying with your peers at a university.
You can definitely study alone and achieve perfect grades, but studying with your peers is how you build relationships for future life and take your community forward as a whole.
I am thinking to make canned encyclopaedia of stackoverflow answers.
Claude/Grok/Gemini/Chatgpt answers are often so… how to say it… misleading? I have to stop the conversation as it leads nowhere (and it is not a skill issue :)
well, SO is probably the highest quality data source for a language model and the rest of the internet is just diluting the final latent space limited by Jon Skeet.
We may need to create a community driven version of SO. Hard for it to be a successful business these days.
https://software.codidact.com/ was created after one of the many SO dramas. It doesn't come up in searches though and I didn't have reason to use it...
Definitely not better off. SO was fairly mean spirited, but nowhere else has such a vast trove of high quality answers to common software problems been collected. SO likely trained many of these models with its answers, and I don't know what software development will look like when it dies.
No where else has such a vast trove of high quality answers been hidden because the question was closed as duplicate when someone else asked that question later.
And, the AI trained on Stack Overflow. So if no one is posting new questions, and new answers. What will AI train on next, for the next thing.
Stack Overflow and Reddit are still getting threads. And as AI gets smarter, the questions will also expand.
your prompts, and the code you have it review.
Maybe.
I thought point was on Stack Overflow, there were community voting on 'best' answer.
If it is just me and the AI. Then the AI training data, is just whatever I approved the AI to do. Just my opinion.
Thanks for this post. Unfortunately, you used the wrong word choice here and this question has 13 other answers that have some of the same words but don't really answer your particular question so it has been deleted. Also, if this remains posted, my not-on-point answer will get less views.
There's more than one reason that forum is dead.
Everyone loves to say this when the death of Stack Overflow is discussed, but it always was that way. Strict moderation, love it or hate it, was part of the platform. And it could have kept going that way for many more years if not for LLMs 99.9% obviating the need for a coding Q&A forum.
Everyone loves to say this... because it's everyone's experience. I stopped using SO as a resource years ago (well before the advent of LLMs) because it got to the point where almost invariably, when I found a post that managed to perfectly articulate my question, it was closed as a duplicate of some other, distinctly unhelpful question. But it wasn't always that way. There's a fine line between strict moderation and draconian moderation, and at some point they crossed from the former to the latter.
Don't forget the other way of sidetracking what you're asking for: "Why are you doing this, do something else instead."
I think most of my questions ended up with this, when I had very good reasons for doing it the way I was doing it. I typically wasn't showing it because I had isolated the problem I was facing into the minimal amount of code to duplicate it, or I was stuck with the particular tech I was using and we had 12 years of code built on top of it and I couldn't switch.
They also neglected to solve the problem of differentiating answers to previous generations of software. How many python2 answers does one have to sift through to get a python3 answer - maybe the weight of answers finally tilted over the probabilities. Even just adding the right tags would have made it easier, but it wasn't ever solved in any way as far as I could tell. And the old answers are there like potholes to fall into.
I rarely posted questions on SO but I largely stopped using it as a resource because of exactly this. I got tired of searching for answers only to find closed-as-dupe questions.
I feel like in their search for “quality” they completely forgot that they needed engagement to deliver value. The whole premise was that the correct answers would bubble to the top, but their system ended up pushing everyone to old questions that had a highly upvoted but either out of date or not applicable answers.
What’s crazy is with the current gen of LLMs you could 100% rein in the abusive mods. SO could reinvent itself as a place for coding agents to get clankermotiomal support after they ‘hear’ ‘WHY THE F DID YOU DO THAT’. (Seriously a hook could draft a submission and then set up a listener for responses. Who would be answering is an interesting question tho)
Ah yes, programming: the discipline which famously only has singular right answers to problems, such that programmers never get in arguments with each other about the correct approach to solving a given problem, and there are no long running disputes that have ossified into intractable disagreements.
Exactly.
Ironically i'm probably a better dev purely because after a few experiences on SO, I would rather waste days/weeks banging my head against problems and learning from them than to actually post on SO. It was a miserable experience generally. For context this was probably ~15 years ago now.
This isn't necessarily to say that SO made me a better developer. Rather i'm just saying that i value (correctly or not) those extremely hard fought lessons. Those lessons where it was considerable pain, effort, time, misery, etc. Are they efficient ways to learn? I doubt it. But in my many trips down that road i developed intuition that i'd probably not have otherwise.
So ironically i guess SO made me a better developer by avoiding using SO at all cost. Conversely, i imagine i'd lack this value that i speak of entirely if i was 20 years younger and starting fresh today. Not sure i'd be better off though.
edit: By "using SO" i should be saying posting on SO. I of course searched and used data found on SO as often as i could. So to that end i am grateful for SO existing.
IMO it was a combination of moderators and users
Sure, the mods were not always the best on SO. But even if you did ask a question, you had to deal with a userbase that was more pedantic and judgy than Reddit. Usually you would get an answer if it was obvious, other times you would have to defend your question against some guy whose newfound obsession was whether you had an XY Problem. Or who was personally offended you weren't using whatever the fad library of the day was (e.g. jQuery).
> against some guy whose newfound obsession was whether you had an XY Problem
Against some volunteer who's encountering their fourteenth clear XY problem of the day.
> Against some volunteer who's encountering their fourteenth clear XY problem of the day.
Fourteenth clear as imagined in their head XY problem of the day.
By far most of the "XY problems" I saw, on SO or elsewhere, were actually "XY problem problems" - i.e. a responder having so limited imagination and character (or, to be charitable, just running very low on energy and focus), that upon coming across a question they couldn't comprehend, they would assume the person asking the question must be confused instead.
That's the thing though, it was voluntary.
If it isn't fun to do, and simply causes frustration, that hypothetical person constructed in the comment could just step away for the day.
I get that dealing with low quality questions wasn't great, but imagine spending an afternoon researching a weird thing using some tools your organization mandates, writing it up, only for that person to skim it and just assume you really wanted to do $otherThing.
This. And it's even starting to be a problem with LLMs - noticed that with Claude and Gemini this week.
Yes, I am specifically asking if it's possible to do X with Y. No, I'm not interested in how to do ${unrelated except for name} thing A with Y, or ${manual variant of X} by hand to ${subset of Y}, nor do I want to use tool Q instead. I specifically want to know how to do X with Y, for reasons that are my own and borne of frustration with Y being a toy I'm trying to use for productive work, which apparently means pushing it past its operational envelope, but I have a deadline...
It's hard for me to imagine a user base more pedantic and judgy than reddit. It must have been really bad.
Could've been a good rule: Unless the XY problem is so severe that X is impossible, you can't heckle or post Y solution unless paired with X solution.
X in an XY problem is almost always possible, like "how do we make ship canals and harbors with nukes" (https://en.wikipedia.org/wiki/Project_Plowshare) or "I'd like to amputate my hand to fix a hangnail".
> but it always was that way.
I don't think that's true. I remember the very early days of Stack Overflow and it felt much more fun and friendly than it did 6-7 years later. I have so many 15+ years question/answer that somehow get revisited by a "moderator" that decides that maybe we should close this.
But was that the cause of Stack Overflow's demise? I agree that it most likely isn't. It's most definitely because of LLMs.
A sink that has large but finite capacity to absorb something can reach an irreversible tipping point when an additional shock happens.
There are many examples of this in nature. (And in Nature [1].) One interesting one that I think is unknown to many people is limnic eruption. A lake can absorb quite a lot of CO₂, for example from volcanic gases. Dissolved CO₂ is invisible, so the lake can look quite ordinary, but the build-up turns the lake into something approximating an unopened carbonated soft drink. If the lake is deep enough and the layers don't mix frequently enough to relieve the pressure, it can build up to the tipping point where the lake will suddenly explode, flooding the nearby landscape and releasing an invisible CO₂ cloud, which will proceed to kill the surrounding life by asphyxiation.
The conditions required for a limnic eruption are rare, though there were two incidents in Cameroon in the 20th century.
It's entirely possible that the build-up of hostility on Stack Overflow were survivable as long as it didn't build up to a level that exceeded the community's ability to absorb it. But an exogenous shock or the community shrinking could upset the balance, with hysteresis making the change difficult to reverse.
[1] https://www.nature.com/articles/s44458-026-00063-5
if early stackoverflow moderation was strict it was strict in a way that was invisible to people who genuinely needed the help. later on it got people who thought the strictness was the main point, and that they had to be vigilant defenders of the purity of the site (wikipedia had a similar malaise).
I personally gave up on the site entirely when I saw a very valid question from an inexperienced programmer closed as a duplicate and redirected to a question about a similar problem that did not actually address what they were asking.
Why are you trying to do that? You should do this totally different thing that I do because I know how to do it.
It was always strict, that was its feature. But it was dying well before ChatGPT came along, due to going from strict to unusably over-moderated.
I'm all for a heavily moderated forum (that's why I like HN) but SO was clearly on the decline (from the graphs in this post) before LLMs came onto the scene. It peaked around 2017 (not counting COVID numbers) and was in a steady decline. ChatGPT just pushed it off the cliff (figuratively and literally).
It was not always part of the platform. Moderation got increasingly aggressive on SO as time went on. There was an inflection point around 2014 where lots of people gave up on using the site. SO was in slow decline for most of a decade before AI showed up to finish the job.
https://blog.pragmaticengineer.com/stack-overflow-is-almost-...
Not to go off topic, but there's some similarities between that and the way that Hacker News is run/moderated. But I believe they've found a pretty ideal balance. Even though they occasionally annoy me with with fun things taken down, I understand the need to have some consistency.
Perhaps they need to take a page out of dang (and team)'s book.
Yeah sure but in the past there was no viable alternative so people tolerated the crazy moderation. As soon as AI offered an alternative people left in droves.
It's definitely plausible that if it hadn't been such a hostile place to ask questions (sorry ItS nOt a Q&a SiTe) that it would have survived AI better.
Everyone loves to say it because it's true. Asking questions on SO was always at best an adversarial process, but it got really bad in the last decade or so.
It was quite simply a profoundly unpleasant "community" to interact with.
StackOverflow's extremely strict moderation was probably the #1 reason it turned into a high quality resource rather than a dumpster fire like reddit.
Yeah, as I understand it, they wanted to optimize for Google search. This meant having "canonical" answers. This killed the site in the long term. In the short term, it worked wonderfully, and the founders made a (well deserved) killing.
Agree. But I asked a couple questions about a year ago and got zero response. It's not just people asking questions who deserted SO, it is also people answering them (probably a chicken and egg problem - reversal of network effects).
This has been my experience. Glad they got what they finally deserved.
if the mismoderation did not kill stackoverflow, it at the least made people who might otherwise have supported the site feel like nothing of value was lost.
The SO podcast was fun when it was running
The author labels COVID and the launch of ChatGPT on the graph, but fails to mention that Stack Overflow was acquired in June 2021 by Prosus, a Dutch private equity firm. That looks to me like it matches pretty well with the entire downward trend.
A firm is sold when its owners believe they will get the best price. The selling itself is more of a symptom than a cause.
It’s not necessarily the sale. Some private equity companies move from “Let’s invest like we’re shooting for the moon” to “Let’s invest like we want to improve margins and flip this on 3-5 years”
It’s not inherently wrong but it is a different model, and sometimes companies suffer as a result.
Businesses (and any other kind of asset) are sold for all kinds of reasons, and trying to time the market to maximize the price is only one of them. Probably not even the most common one.
> Stack Overflow was acquired in June 2021 by Prosus, a Dutch private equity firm.,
That is great to hear. I am glad that the original creators of StackOverflow got their liquidity event and are well off financially I suspect.
What did they change?
1.8bln
Stack Overflow with all of its shortcomings was a marvel of the internet at it's peak. People especially in early were chasing karma and anything you asked, you were sure to get some answer. Not always right but some answer. While for sure LLMs will give much better answers on average. I feel that it's a piece of humanity we've lost there that should be adequately remembered and the memory cherished.
getting a wrong answer in a public forum can be great to motivate corrections :p
Wouldn't this be worrisome? People used StackOverflow and generated new knowledge along the way. Without such medium for discussion, how can we feed models with up-to-date quality knowledge?
Plenty of documentation, and plenty of code that the AI can read itself.
E.g. if a library has a bug that has a common workaround, it can learn that from open source code using the library that uses the workaround.
I don't think its much of an issue
- Rl envs + synthetic data + human annotated
- Usage data from codex/claude code/cursor
Most of the model abilities in coding come from post-training, not pretraining
A better question is what's left for those who don't have access to that. We went from publicly available to vacuumed from private users
Open source models
unfortunately all the incentives right now are for repos to be private
Careful, you can't point out that the AI emperor has no clothes or you'll get called a Luddite.
How do you convince people to not want an instant answer? Even if SO didn’t result in so many “What have you tried?” responses and immediate closures, most people would still prefer instant feedback.
Pointing them to docs? Which is anyway what stack overflow answers did?
I wrote multiple answers to questions that weren't just "point to docs". And even when it is pointing to docs you are providing the reasoning as to why it works one way or another.
What docs? Who writes docs now that AIs answer everything?
Ever since the AI stuff started rolling around on coding i've seen MORE documentation, theres a big incentive to properly document your API endpoints so LLMs can figure it out from specs, and even when not documented the llms can also just read the code and figure it out directly (for libraries and similar). And at least in my experience they tend to document or write it down for future sessions too!
I know you're being facetious but there may well be docs. It's just that the same AI most likely wrote _them_, too.
Did anyone (person or competing LLM) bother to verify that they're correct, though? Who knows! Let the next generation of models worry about that.
I've heard this is now most of some CS jobs now. Just writing documentation for AI.
on the contrary, theres more of an incentive for apis to have docs for agent discovery. the docs / interfaces themselves can be auto-gened (stainless / mintlify)
We unironically need an StackOverflow for LLMs.
LLMs would post solutions to the issues that they've discovered after doing a lot of research.
Unfortunately the LLMs are concentrated into few providers (OpenAI, Anthropic, Google) so there's a chance they each end up doing their own private (and closed) StackOverflows. By leveraging their private StackOverflows, their LLMs will be able to short-circuit complex reasoning, saving tokens, time, and money.
Yes. Very.
I knew that stack overflow must be suffering because of AI, but I find it hard to believe that questions asked per month has gone from 200k pre-chatbot to (what appears to be) ~1k. Although, I suppose I have not gone there at all in the last 4 years...
Clicking through to the query for the first chart, I see the peak of ~300k in May of 2020, and it was ~3k in April of 2026 (the last complete month). I’m flabbergasted.
https://data.stackexchange.com/stackoverflow/revision/193252...
Stack Overflow might be the greatest receptacles of human knowledge on programming.
But I would argue that it usefulness only extends to its body of knowledge. As a service and/or community it has been pretty terrible for a long time:
If you were a new user trying to learn programming, it was maybe one of the most toxic resources available. I don't think I have posted a question since 2019. And even there, the only thing the average user could expect was a snippy response from someone who barely stopped to read your post. And/or a mod deletion because a similar-ish question already existed (regardless of whether it had a satisfying answer).
At a certain point, all the meaningful questions have already been asked. The site exists to collect novel new problems and not help people with iterations on existing problems.
(Also, underrated is the extent that the industry has homogenized around a couple of frameworks that are used for everything. I think it's telling that the peak of StackOverflow coincided with the era that React was taking off, to just name one).
Early years SO was optimized for people helping people. Later on they ruined the site by optimizing for tidiness ... and griefing users (especially new ones) off the site in the process.
LLMs are better than slow human support of any kind for debugging / helpdesk work (which was never that welcome at SO anyway).
Stack Overflow is still great for canonical questions, multiple answers, public / SEO'd discussion between humans, etc.
But that probably isn't enough to save the company as a private equity acquisition hoping to 100x their $1.8 billion investment.
Hopefully the classic Q/A site eventually gets written off and spins into a Wikimedia-like foundation that is interested in preserving the original Q/A site and has no desire to grow or become something else.
This is happening to Reddit too, albeit in a different way. Almost every other comment on popular subreddits is from surreptitious LLM bots.
I feel reddit is having a near death moment.
There are prowling bots trying to strike up engagement with stupid open ended questions "do you find that using a golf simulator improved your golf?"
And some subs seem infested with submarine advertising, posts that mention a single product name almost in passing.
Nearly always these people have their posts hidden. Reddit has always been looser, people can edit and delete their comments and entire posts, and enjoy some frothy conversation while hiding their old rants.
There are plenty of signals that reddit could use to push out bots but they just don't seem to prioritize it.
When you find your self wasting time responding to a bot it's a bit of a sucker punch. Too many of them and Reddit will be on the ropes as a wasteland.
Just look at the graph, Covid peak aside its previous peak was in 2016 and it was in continuous decline since then. All LLMS this was increase the slope.
My speed was already declining on the highway after I let off the gas, all the brick stopped truck did was increase the slope.
Forum? What forum? When has Stack Overflow ever been a forum?
How is it not a forum? Sure, it has some search features and a "comments/answers" dichotomy, but at the root it's just phpBB with fancier formatting
Wow. It declined all the way to zero? I'd have expected it to tail off.
That's scary. What else can AI make decline all the way to zero? Customer support?
I know everyone here is heckling calling it a forum, but this is a basic forum feedback loop -- the forum activity declines, so people show up to check less often and get fewer responses to their posts, so the post there less often, repeat until traffic falls near zero.
You may have like a handful of weirdos who never leave and develop their own little community in the wreckage, especially if the cost to continuing to run the forum is trivial, but it's basically a death spiral every time.
> You may have like a handful of weirdos who never leave and develop their own little community in the wreckage
Slashdot is firmly in this stage of its existence, and honestly it's kinda fun
one of the interesting properties of public forums like SO is that the maintainers of software packages get visibility into the problems and frustrations that their users experience
we're losing that signal when the Q&A behavior shifts into language models
Did the founders have an exit or liquidity from Stack Overflow? I hope so.
https://blog.codinghorror.com/launching-the-rural-guaranteed...
Wow, has it really gone all the way down to zero?
i haven't been on stackoverflow for maybe 6 months, haven't seen it on google search, or needed to ask much on google either. Maybe they did a web search. On the plus side, its still used by AI agents
StackOverflow always sucked. I never found it to be a useful resource.