Stack Overflow announces 28% headcount reduction
stackoverflow.blog
stackoverflow.blog
Yeah, as someone who moderates a 200,000 member community, moderation is difficult. They do a pretty good job considering what amount of shitty, low value, googleable questions they get.
What do you all do if ChatGPT's prices rise? Or you're using a technology it knows nothing about? Back to IRC channels and reading docs? Where will you copy paste your code from? Cling to it and hope an alternative comes along, all the while shoving more money into closed models?
Working it out yourself is always going to be a good way, IMO
My main use of SO is figuring out badly documented DSLs or configurations or magic annotations that have nothing to do with programming per se.
I can dig into a language I don't actually know and change a value or fix a trivial bug, as long as it has variables, methods and loops.
But I can't guess if @Zorpify is the magic incantation on a method that does something I don't know the name of.
Check out 6:05 and 6:25 in this Andreas Kling video: https://www.youtube.com/watch?v=8mxubNQC5O8
That's where Copilot shines, not for locating and copypasting answers from Stack Overflow.
Pay for your tools? This is not exactly a community of low-income people here.
I think the real benefit of local models lies elsewhere.
Especially the people who need a better Stack Overflow, are likely not to have huge employee salaries.
The prices of such tools are often oriented towards people who have high-paying job in Silicon Valley. Not everybody who reads or posts on Hacker News belongs to this group. :-(
Discord sucks as a community for open source/software development for more than those reasons. The major one, I find, is the population of gamer/anime users that dominate it - I feel alot of serious developers stick to slack or mailing lists for that reason. Slack's identity is clear - enterprise/business chat and communication. Discord's is a bit more confused and alot of people straight up haven't even heard of it.
Yeah see this is the problem. Fuck that man. The vast majority of people who want to ask questions are very new. They barely know how to phrase the question they’re trying to ask. If they knew how to ask it, they probably wouldn’t need to post a question. They could search for it. SO is full of elitist pricks who are very rude to people who are trying to ask a question and don’t know how and are feeling frustrated. It’s a beginner hostile environment. And that’s not intuitive. Because you DO end up hitting stack overflow for a lot of basic questions from 10 years ago, and that’s nice, but it is bad for people who lack the expertise to navigate that historical index.
Discord is not indexed, but it’s largely full of people who are willing to help newbies construct their question and then answer it. It’s night and day a better experience. Same for chat gpt.
I have spent a fair amount of time on the unreal engine discord helping what I assume are kids who have never done any programming before realize that their questions make no sense and that they don’t know what a variable is. They are extremely thankful.
SO just makes people feel bad for not being experts on the topic for which they are seeking help to learn.
this is a big one. often you know what you need to do, but knowing what to search is hard.
Half of the time just thinking about trying to ask a good question will usually lead you to a solution, or if nothing else if you're not finding an answer through searching you're probably just asking the wrong questions because the odds you're solving some novel, undocumented problem is pretty much 0.
So usually it's more of thinking one knows what they need to do, just to learn that they didn't actually know what they needed to do at all once unknown unknowns are unearthed. That's where something like Discord shines because it provides an avenue where extremely patient people will do some prodding of a newbie to help figure out what it is they're actually trying to do so they can be pushed in the right direction. But that's also a double-edged sword because searching Discord is extremely difficult and has a high barrier to entry so you're often going to be faced with the same questions being asked over and over.
Besides that, the answers you do find are often not helpful. For example, “just use this function in jQuery” - I’m asking about JavaScript, not jQuery.
Since jQuery is open source, I do consider this to be at least a helpful step, since you can then simply look at the source code of the respective jQuery function.
But I doubt this is a value prop people want.
Edit: sufficient to sustain SO in light of new competitors
They got so uptight about moderating content that they seem to have forgotten their actually user value. Instead of focusing on helping users find more relevant content, they’ve decided they must reduce any possible duplicate or overlapping content.
For me, I’ve found most of my issues are complex and nuanced. I want the duplicate content. I want to read 5 posts with slightly different variations of the same issue. They all provide contextual clues and variations that are useful for solving my specific issue. Instead, they all get forced into the same post, resulting in a bunch of absolutely useless junk. Top that with the fact that some information is actually stale, it’s basically impossible to find a solution now.
ChatGPT often gives me solutions that are mostly correct. Even when they’re wrong, they often get me close enough that I can read the docs.
I don't think you can say this is a community with opinions. Rather, there are various subgroups of people with strong opinions and certain posts attract one subgroup and certain posts will attract others. You can find a whole lot of people telling you building an app "on top of" an LLM is a dumb plan because they aren't reliable. And you can find a lot of people telling you to delegate all your drudgery to an LLM. That might seem a bit contradictory but it's really slightly different topics pulling in slightly different people.
And in the this case, there are lots of people willing to tell you bad things about SO. The story for such bad feelings is that SO was fun to do answers for and now it's really miserable. It may or may not sometimes still be useful in an objective but it's accumulated massive bad feelings. So just about any SO thread attracts the SO-hating subgroup rather than any other subgroup.
If I asked Stack Overflow a question, they made sure I wished I hadn't.
I refuse to interact with the platform after these encounters.
All I was doing was just: putting effort into my questions
A lot of questions show people have not done that by either reading manuals or searching SO.
Why should a answereer put effort into writing a good answer when the questioner has not shown any effort?
My understanding of the reasoning behind some of the moderation - flawed as that often is - is to make the questions and answers 'general'. Which means a question that is very specific to one person may be voted to close or voted down.
To take an example of another stack exchange site (chemistry.SE), they get a lot of 'homework questions'. Literally some questions are like "I have an exam in an hour PLEASE HELP". The stated policy is _not_ to just answer these questions, but to guide people towards a solution. Very frustrating for the person asking the question - who, you know, has an exam in an hour ...
Ultimately there are policies and processes intended to improve the overall quality, and sometimes that is a good thing and sometimes it leads to everyone getting annoyed with the whole thing.
That's the opposite of the guidance, or at least the guidance as it was back when I was active on SO.
General questions "What's the best way to do X" are shut down.
Questions should be specific and about actual problems you have rather than general problems.
Solving real problems with real solutions builds up a strong data set, not questions about imagined problems and their answers
Ultimately though, Stack Overflow just faced the customer service curse: as volume rises, moderation gets blunter.
May I ask your SO id? It'll be a link like https://stackoverflow.com/users/1, except with a different number.
Consider two scenarios : a) I have installed some software X but messed up one of the install steps and now there is this weird error. Or b) when installing this software, it always errors if library Y is on an old version.
For b) it is clearly a 'general' problem that many people will face. For a) it could be seen as a user error and therefore not generally useful. From the point of view of the people facing a) though, it absolutely is a problem!
"How do I sort a list of US States in Python 3.10?" Too specific.
"How do I sort a list of strings in Python?" About right.
"How do I sort a list with my computer?" Too general.
A great SO question ① is about programming, ② has a factual answer and ③ if other people need that answer, then their search terms match the wording in the question. It's that simple. Answering that kind of question is worthwhile for people like me, because many people will later search, find the question and benefit from the answers. Spread the knowledge!
And I disagree that questions need to have one factual answer. Great questions often have many highly ranked answers showing different ways to solve issues or emphasizing different aspects of solving the question.
A great question is one which others also have.
2. It sounds as if you think the moderators and the answerers are disjoint sets. They're not, at least not in the tags that I follow. The power tripping mods are the same people who contribute a majority of the answers. The decision of whether to answer or close is one decision, not two decisions made by different groups of people.
Mostly, moderators and answers feel like disjoint sets because the majority of answers come from the long tail of people (at least it feels to me).
But likely it is similar like with Wikipedia where the moderators feel like that they do most of the work when in reality they really don't and others would be happy to do the work if the moderators wouldn't hamper them.
There have been exceptions, e.g. we asked a site moderator to wipe someone's bas64-encoded password from a question. The user probably changed password, but we got a site moderator to wipe it thoroughly without waiting for that. I know those site moderators do a ton of work, but the power trip behaviour you see as questioner is done by answerers like me.
The site moderators do things like cleaning up vandalism or spam, when someone uses a script to add outgoing links to a thousand pages.
It'd only get worse in cases where only a human had access to the knowledge or the means of attaining that knowledge.
Besides, Github, Gitlab, Bitbucket, etc. have far more public data to train on, Stackoverflow only really provides the schema one would want for fine tuning to make sure the answers are in the right general question-code-answer-code format.
Once that is trained properly (see: to the understanding level of compression, not just memorization) you only need to update the model on new libraries and languages, which can be trained from repos and docs directly.
They ran adds to cover the servers and development cost. A fork must find how to get money.
Besides, Stackoverflow runs ads and profits directly from users' posts. You don't get to own things just by the virtue of being a platform that lets people host things. Does Github own all the people's code they host? Does a cloud service own the software people put in their containers? Of course not. Stackoverflow themselves state that each answer is the property of the poster, licensed as CC BY-SA. Even Reddit's TOS states that "You retain the rights to your copyrighted content or information that you submit to Reddit".
Now whether there should be a class action lawsuit including everyone who's ever posted on Reddit or Stackoverflow against OpenAI is another story.
- Ask ChatGPT (or a similar AI chat bot) a question, it at best gives some at most semi-true answer. In the best situation, I can immediately see myself where the mistake in the answer is.
- Look at Stack Overflow: very often find an answer
- In the rare situation that I cannot find an answer on Stack Overflow, and ask a question: get a good answer (often many good answers) after a few hours that let(s) me realize that I have a lot to learn about the respective programming topic, which I will do.
- from 10 years ago, and an entirely different version of the language/library
- from more recently, but it's been closed as a duplicate
- with a specific caveat, but that guy ignored and the question was also closed as a duplicate.
- with a wrong answer, that you'd like to post a correct answer for, but you haven't karma farmed enough
Is there something about stackexchange format that accidentally encourages this? I go on various discords (learn piano, javascript, typescript, various games) and the absolute worst treatment I got was nobody replying.
https://files.catbox.moe/l5n5gd.jpg for some of the results in their technical report
Which says more about your style of asking than about the SO community.
You could try a "I know I should be doing X but I want/need Y" disclaimer (also often useless) or spend writing half a novel, to only get told "can you make that a shorter repro?" or back to square 1, "you shouldn't do that'.
Most of my reputation gained there is because I answered my own answer and ironically - just once - copied a solution from a thread on GitHub and pasted it there.
Had I been a minute or two later I suspect it would have been closed as poor quality.
I only spotted it by pure chance because I had just wondered the same thing before having a "duh" moment when I realised it was a really simple solution to what sounds complicated (but isn't).
So I was able to interpret the "real" question and answer it.
But, as you say, never a bit of feedback from the question asker. No acceptance, just abandonment.
It just slowly gathers karma, around 10 per month for ~10 years.
[1] https://stackoverflow.com/questions/27607516/sql-join-with-c...
Instead of anything that, they let mods use a bullshit sniff test, then changed their mind on that and got into an embarrassing war with their mods. Obviously there's a little bit of empathy to be had for the mods in terms of dealing with generated answers, but the answer of banning them outright seemed very kneejerky and a plainly bad idea and entirely ignorant of where the winds were blowing. An awareness of how the mods were responsible for turning SO into the... interesting place it was at the time and being prophylactic about fixing the culture they had created towards the end of the 2010s would have helped prevent this.
Instead here we are, a year removed from the beginning of this and it really feels like the Stack Overflow is a much, much smaller part of my life than it once was. Sure, in a few years the training data for GPT will be worse, but it's not like I don't still find Python 2.x answers when clicking on SO links, and I can paste in the new docs, and not worry about whether some random moderator is having a good day or not.
It's sad, really. In time we'll see Stack Overflow's decline as preventable (at least to an extent), and view it as a case of what happens when the strong leadership that's needed to disagree with a loud minority is absent. Whatever you think SO should have done, I'm sure that you think what they did end up doing was not what was the best play for them.
I imagine that people are now going to demand a premium to work there - they might only be able to attract the kind of people that want to treat it like contract work and expect a premium for insecure work to cover their expenses while they look for the next contract. I imagine this will put the company at a significant disadvantage compared to where it was before. Finding any marginally profitable use for their surplus employees (even if it is building a different product or even taking on some contracting) while they wait for the workforce to shrink through attrition would probably have been better for their company long term.
It's the same behaviour of course, blindly copy/pasting errors into ChatGPT, unquestioningly copying output back, and as far as I can tell the results don't seem to be that much better, just quicker.
Sometimes it's a 1:1 copy of stackoverflow, sometimes it's a bit more. Either way it saves me a few minutes searching for the relevant post.
It beats searching various SO questions, some of which don't have answers, some are too old, some are hard to read.
The best part is being able to ask for further clarification, or feeding the answer into a fresh context chat to help find problems
This is actually a shame, as lots of very knowledgeable and helpful people have answered questions over the years on SO, and sometimes the comments and discussions are just as useful as the accepted answer.
After a couple of "you do know nothing, go play over there" replies, I stopped visiting and participating in SO, and even started to avoid going there because the answers' quality started to go down as well.
They tried to solve it with a CoC, better moderation, etc., but they only slowed down the process a little. In my eyes, it became a free Experts Exchange filled with people with short temper.
If I knew better, or my half day search yielded better results, I wouldn't be asking with all that I know, in that poor form. Judging me because I'm not as knowledgeable as you is not a good way to help people, it seems.
...but, mostly not.
Mostly, they're quite useless. ...and I say that with a considered opinion, not a knee jerk of irritation.
SO is fundamentally a site where up voting lifts relevant answers and down voting sinks irrelevant stuff.
However, comments are a side-step into non-voting, non-karma related content, that violates that fundamental premise. Comments simply give people a way to game the system; you can say unhelpful useless things, and the worst thing that can happen is the comment is flagged for moderator review. You can answer the question without it being a 'real' answer... only now it's harder to find.
The benefit of comments is entirely lost by the way they are abused in practice.
...so, realllly, if you want to chat about the solution, then chatGPT is the way to go, not SO. SO was never the place for that.
So.. you know. It's not much of a loss really, compared to what you get with a chat bot that's happy to talk to you patiently asking dumb questions literally forever. :)
So it can't replace these sites in their entirety, because if those sites decline, so will the quality of ChatGPT responses. I suppose there will be some kind of mediocre equilibrium.
1) Nobody thinks that Stack Overflow is a magic oracle. Most people do think that ChatGPT is.
2) Quality of Stack Overflow answers can be quickly judged by looking at upvotes. Lots of upvotes does not guarantee that an answer is quality, but it is a signal that is very useful especially when you aren't capable of judging the answer yourself.
The anal retentive people that we complain about, are the same ones who fed the AI. When they don't have a home to torment us, where will the AI get the answers? Training on code is "cute", but not enough.
Sorry to anyone from SO reading this, but I have been using the site for.. probably 10 years (or how long it has existed) as mostly a reader and there are 2 features. Have the answer to my google search ready. (It's two and not one because I also need to find it)
Why the hell does this company even have so many employees? It's a forum and I've not used any new feature. Person A asks questions, Persons B-Z answer. Someone fixes the bugs, updates the dependencies and someone else keeps the hardware and network running.
I suppose it's me. I don't need exciting new features, I don't need growth, I don't need an experience. For my needs SO could be run by 3 dude(tte)s in a basement, and as long as it's profitable enough for them, all is fine.
Autonomy so they dont get a do nothing manager that wants meetings to discuss upcoming meetings.
Treat them like adults - also hire adults in the first place. Throw out well poisoners early on. Bullies, sexists, racists, over-competitive twats, ass kissers, boot lickers, etc if you somehow hired such types - throw them out too.
Pay must be market rate but doesnt have to be 500k for an api plumber.
Remote work so they buy a property they afford, settle down and raise a family.
And so on.
This is in opinion a very US-centric observation/expectation.
If you continue reading, he doesn't mention the community as the most important part of the "we" at all.
Added as a side-note to not be so obviously greedy:
> To our community members and customers reading this note: you are foundational to our success.
Great Work Stackoverflow executives!
Just “employees” would make it sound less tone deaf.
"Employee" has come to mean transactional and disposable, and I don't know how many leaders still feel obligation towards employees.
Maybe thinking of people as people, rather than as what "employee" has come to mean, will activate feelings more like "our people" or "my people".
There is a big message telling users not to use AI generated content for answers
I'm starting to wonder if the no AI answers rule is a good idea
Most SO answers tend to be near the correct way as they explain why they answered.
A good answer should also gove references as to why the answer is correct.
The gamification of the Stack Exchange sites is a double-edged sword, and I think it relies on an assumption about the effort threshold required to provide an answer.
Typing out an answer, even a wrong one, takes time and effort that the system then rewards through points and badges.
If you can just copy a question to Copilot and then paste the answer back, the effort threshold is likely then low enough that too many (or at least lots) of people will do only that in order to engage with the gamification and get points, rather than do it to try to help out the asker.
There were too many questions asking the same thing that had been answered many times before.
Too many where the answer is RTFM.
SO used to have interesting questions, questions that I needed answered or that I found interesting to spend time on to research.
I think that is precisely the problem. A kid is teaching himself to program. He has an error which a million people before him have faced. He types it into google and gets SEO bullshit. He asks stackoverflow and they close his question immediately as a duplicate of a question which, to him, looks very different. He asks ChatGPT and he gets an answer immediately.
Where does start next time he has a problem?
So it will be with the possibly wrong answer from ChatGPT.
When learning nothing replaces a book or tutorial that has had some proof reading or editing.
Just picking up small bits from the web here and there does make things difficult and does not really teach you.
Learning in general needs you to fail and think not just google an answer.
IMO a lot of the interesting questions got moderated away. Almost all of my favourite SO questions are closed as being off topic or subjective. I do think that the community struggled to scale with influx of a huge volume of users. But I also think that the strategy that they chose to deal with it was ultimately overzealous.
Your definition of interesting is a discussion which is not what SO is for - even if they did then they do not have to tools to help this e.g. threading being able to vote on interim comments etc.
Even if it is it is often difficult to put the individual facts together and that is what a SO answer will do.
The issue is that current AI answers invent 'facts' or produce wrong code. AI might help improve how people questions and possibly help them find an answer e.g. improve searching but to answer you must be correct and give correct references - not false ones as have been found in legal cases and others.
How to use AI is an open question.
You need cookies to count the views.
I love SO, but I really wish it would have become some community funded / non profit entity. Who wants to help these private equity firms?
Is it cynical to wonder what "successful launch" means in this context, or just HN being overly dismissive or out of touch?
> As we finish this fiscal year and move into the next, we are focused on investing in our product. As such, we are significantly reducing the size of our go-to-market organization while we do so. Supporting teams and other teams across the organization are impacted as well.
This means they're cutting down on new product introductions, and are focusing on generating profit from their existing/deployed products?
If someone can offer some insight on why we're paying somebody to spend a week thinking about what to call the layoffs ("headcount reduction" lmao), and not continue to pay the actual contributing members of a software team, and especially what the angle is in writing a blog like this instead of quietly taking care of your people, I'm all ears. Is it marketing trying to justify their continued existence? Are we helping the people we're letting go by announcing the flood of (probably experienced) talent into the marketplace? I'm only disparaging this practice a little bit. I genuinely am curious about what the psychologists and market analysts have found about how this is beneficial to the company.
[1] https://www.wsj.com/tech/ai/ais-costly-buildup-could-make-ea...
I've been following layoffs for the last few years, and it generally seems like ~10% is normal, with larger companies doing a smaller percentage, while startups often layoff 10% or slightly more.
28% is a scary number.
I'm betting that everyone on this site has visited StackOverflow at least once. If they cannot figure out a way to weather an incoming storm, then this makes me even less hopeful about tech employment in general.
But, then again, SO did get purchased in 2021 by Prosus (https://en.wikipedia.org/wiki/Prosus).
Scary times ahead? Or, just more evidence that private equity ruins everything good? Or, an entirely new story that AI is going to rapidly destroy all businesses?
We are coming out of a ZIRP (zero interest rate policy) period where debt was effectively free and lots of companies borrowed and grew very quickly (some/many/most probably knowing they might eventually need to shrink back down some amount if the financial climate changed).
I have no idea if this was the case with Stack Overflow, do we know how many people they hired in the last 3 years, leading up to this cut?
A common trend in these lay-offs is that I can't for the life of me come up with what those employees would be doing. That combined with a sudden lack of easy money due to rising interests rates and more uncertainty about pretty much everything seems like a simpler explanation for these lay-offs.
I think it is this. Private equity translates to layoffs regardless of the need for layoffs.
ChatGPT 3.5 started giving me wrong or not so good answers. So I rarely use it these days.
(I'm thinking contributed directly, not indirectly, from all the copy&pasting answers into Git repos.)
The way things are going, there’ll be islands of content and AI trained on it.
Like how Reddit and Twitter closed crawling of data.
https://www.businessinsider.com/stack-overflow-acquisition-p...
Whew! I was really worried that selling out to private equity was going to hurt the company in the relatively-short term.
Seriously, American capitalism is increasingly persuading me to be a socialist.