I don't think I've ever seen a post on SO with that many downvotes before. That's pretty telling how the community feels about being used as training data (especially backed up by the highly upvoted reply saying such).
That's nothing for a meta post with an unpopular announcement, there are quite a few with hundreds of downvotes. But yes, this does likely indicate something about how the active SE users feel about this. It doesn't help that the blog is written in a way that is much more likely to appeal to shareholders than software developers on SO. But ChatGPT has been a quite significant moderation issue, and the blog post doesn't address any of this (it is kinda devoid of content in general beyond "SE will do something with generative AI, updates later this year").
meta forums on SE sites see disagreement a lot more often than the regular answers areas. These are usually longtime SE users with a minimum of site credit to participate. So while these are are not the bulk of SE users their voices should ring louder imo.
have I been missing something all these years? How do you see reply votes? I only see # points (not actual up/down vote count) on my own replies, and the OP.
Click on the votes themselves. This is at 1000 points as sort of a rate limit so that only a fraction of the site can do it) and to ensure that people understand the nuance of voting better.
"That's pretty telling how the community feels about being used as training data"
I don't think most of the replies on SO mention this directly, they mostly focus on how this will impact the site's quality negatively. But really you are correct that people don't like being used as training data and imo the "arguments" everyone are making against AI disadvantages on SO, on art sites, on wherever else are pretty much "excuses". Quoted because even if they are correct they are just excuses for the core much more ethical issue - humans don't want to be replaced. Being replaced by a machine is even shittier. Being replaced by a machine that exploited your own hard work against you even more so.
In a real world example if my boss hired a junior whom I must mentor with the sole purpose of me being fired and replaced so he can pay him less. Well that would be considered very unethical even though it does happen and many workers do put up with it. Yet the current AI movement is that on an unprecedented scale and to the people being replaced it must feel like a massive bitch slap in the face.
To me it seems like an extreme expression of power and it's bizzare how many people are fine with it.
Years ago. AI bots like ChatGPT will make SO almost entirely worthless (worth so little there won't be enough demand to sustain a large service, it'll collapse except for being an archive).
The fact that this post by Prashanth Abd al-Rahman (SO CEO) is so downvoted makes me think he's onto something. This is like when artists cry because of how good Midjourney is and they'll have to switch careers. If programmers are complaining about AI it's because it's good.
This is a bad heuristic, strengthening your priors based solely off rejection.
Nobody questions LLMs can write code. The question is whether Stack Overflow has a place in that future. The community’s rejection, and CEO’s ham fistedness, suggest the data he has are the data he’s got. Which makes him uncompetitive vis-à-vis e.g. GitHub or any IDE.
No, the AI can only replicate what some other human figured out before and wrote down. Right now LLMs aren't really creative problem solvers. Therefore he needs a bunch of idealistic/stupid/unaware devs to keep sharing novel solutions that can be used to improve whatever AI he wants to have and sell it to companies which will fire those same devs that figured out those answers.
Isn't it obvious that the new AI gods will require our regular offerings in the form of text, pictures, music, and anything else that can be manifested in this existence we call reality? Create, and praise the transformers.
Transformers and AI will discover that 98% of human expression is just a rehash of something else. That 2% of original expression will become 50x as valuable instead of getting lost in the noise. This may lead to unpredictable disruptions in the extreme regularity of human behavior as people start to realize how boring and unoriginal everything they've ever thought and done is.
"as people start to realize how boring and unoriginal everything they've ever thought and done is"
Most people actually like stability and no surprises, when doing their habits, like drinking coffee for example. It should not suddenly turn to wine, that would freak them out.
And books and movies should follow a familiar pattern, or it will be perceived as disturbing. Slight variations are welcome, but nothing too drastic. So I doubt people will be truly shocked.
That would be a hilarious future if humanities job was to upload our creative offerings to AI and it dispenses some trivial amount of crypto based on how much it learned.
Your 8 year olds cat drawing has been subsumed for 25c. Your essay on Philip K. Dick has been subsumed for .32c.
seems folks are mad because Stack exchange ceo wants to use their own LLM to provide answers on their site. Which is something that other SE sites have banned as answers previously.
the use and abuse of the term "community" to mean things convenient to for-profit content aggregators. What could be wrong with the New Digital Feudalism?
The arena is saturated, but the quality is garbage. Lots of 3rd party recruiters hiding company names posting the same job everywhere.
If you can keep the quality high for a small-medium group there is money to be made, I imagine (i'm trying to start a job board myself for a tiny group). But just look at anything Pieter Levels has built (remoteok.io) dude has a huge monthly MRR. granted hes a solo founder and SO probably employs many. Also look at Dice.com. They allegedly bring in near 150Million/year! To have a targeted audience in the palm of you hands and throw it out the window is just bad business. SO is really just throwing away money.
Maybe SO management figured it should be more lucrative to have those job sites/brokers bid on ads placed on SO rather than becoming a competitor to their customers.
I used to work for company helping with hiring process, we had integration with most of them.
The consensus at the time was that end to end wise (company hires and is happy with candidate) - you could just remove every integration leaving just stack overflow jobs and it'd be fine.
It's beyond me why they cancelled it, so stupid from my PoV.
SO was phenomenal for sourcing engineering talent. No board I ever used since gets close to the level of incoming candidate quality. I am still sorry it's gone, and I'm especially sorry every time I need to interact with LinkedIn hiring instead.
I loved looking for jobs on SO. I found some of the coolest companies and was seriously bummed when it went under. It’s how I heard about my current job I’ve happily been at for a couple of years.
Prashanth Chandrasekar was brought in as the hatchet man to facilitate the sale after destroying Rackspace. Fired the beloved community managers and the community became much more hostile since. Why are any of you surprised he is out of touch with the community?
SO should be ran by a foundation funded by tech companies enlightened enough to realize the massive productivity gains it produces (with SE as a side goodwill project). In this era, this counts as wishful thinking. Instead we got a 1.8B sale, that's not pocket change, now the profit must flow.
Only one really, this being Internet Archive. Mozilla [1] and the Wikimedia Foundation [2 - talking about Wikipedia specifically] are too ideologically tainted to be seen as "enlightened". I have not heard of similar problems with the Internet Archive so I hope that sanity prevails in that organisation so that internet history is recorded regardless of the ideological bent of what is archived.
In addition to the problems you listed, Wikimedia spends money like it was water, and Mozilla's handling of their main product, Firefox, could be charitably described as mismanagement. Neither organization deserves trust.
The Internet Archive that just risked their existence on a why-did-we-think-we'd-get-away-with-this blunder (charitably you could call this ideological blindness, at best?)?
But it is possible for tech companies, even ones we generally think of as evil, to collaborate (entirely out of self interest!) on mutually beneficial advancements.
Don't mistake "it would be wise for tech companies to support such a thing because they would get ROI" with "any company who supported such a thing should automatically be declared noble and perfect and of unimpeachable character forever"
Once upon time the name Rackspace meant quality and the utmost in customer support. You paid for that, of course.
In 2015 Prashanth Chandrasekar became Vice President and GM of the company -- the leader in all but name. In 2016 Apollo Global Management bought the company and took it private. The rest, as they say, is history. Sad history. And yes, he is 100% to blame. Or, from his and Apollo's perspective, to praise. Gutting a company this big brings in a lot of dough.
It took him two years to sell Stackoverflow. It's a harder sell, after all, than Rackspace.
Providing this is opt in i'm pretty OK with it, unless i'm missing something?
We trialed a private SO instance at work, and discoverability was an issue (now we have to search slack engineering channels, check notion, and check SO).
If I could quickly ask, in for example slack, a question in natural language and have it query our private SO, slack, notion etc, I would be pretty game (providing this data remains private).
SO is a toxic cesspool, especially for people new to programming or just looking to learn. If ChatGPT and other AI tools can get the job done without resorting to asking anything from the SO “community”, that’s a victory.
For me, both. I search for something generic, either an official manual, a bug report or a StackOverflow's answer pops up. I have a specific narrow question, I'm unable to google the answer, I ask, a more knowledgeable person appears and ties all the loose ends together in I way I have never thought of.
The problem is if you solve the problem of not needing the community by training an AI on community content, your community will leave and then you're out of training data. AI is impressive, but I'm unconvinced it can answer truly novel programming questions.
ChatGPT can not do anything especially not in programming.
All it can provide is how the answer would sound or look like.
Be prepared for an absolute avalanche of bugs and security holes no one alive would've made. I make my living from debugging so I welcome the new job security but it's a massive net negative for society, no question about that.
I asked it to write a simple function to copy text to clipboard. It unecessarily created a async function and forgot to pass down the text as a variable.
Next, it tried to use AlpineJS to create tooltips with NextJS which is really not supported and couldn't fix the resulting bugs.
For sure, it's better than nothing, but you've got to be at least somewhat competent before you start copy-pasting code from it blindly.
A bunch of the comments seem to be of the opinion that "using LLMs in S/O" === "LLM answering questions"
But that isn't the only task available for fancy auto-complete. Eg. The LLM could help novices make better questions, or include more/less context in an answer, with the person still being the arbiter of truth.
I have used LLMs before to understand broken English. Being able to read someones words without stopping every few lines is a major benefit to interlingual communication.
One of the biggest frustrations for people on SO is struggling to find a question that already exists - and then people get really angry when their question is closed as a duplicate... but what if the LLM could take proposed question text and point the person at an answered question so they don't even have to wait for an answer or a duplicate? It'd be much better than the current duplicate finder.
>but what if the LLM could take proposed question text and point the person at an answered question so they don't even have to wait for an answer or a duplicate?
Has there ever been a case of a company the size of SO deciding that, given their skills and culture, they’re just going to wind the company down rather than try and compete in a new technical arena that they’re entirely unsuited for? I’d respect that.
I don’t know if what he’s suggesting here makes sense (I tend to think no), but I’m automatically a little skeptical of the typical response to a serious threat: “no way, y’all, this is actually good for us!”
Imagine google coming out right now and saying the future of AI is search ads.
I would suspect that the company is too small to train their own LLM from scratch. But Stack Overflow probably has too much traffic to just pay for something like the ChatGPT API and built something on top of it. I'm not sure how many good options there are in between, can you realistically create your own LLM for this kind of specialized area without the kind of resources OpenAI/Google/Microsoft have?
Online discussion happens at the edge of what is known and that is where AI learns from. Yes, humans talking in a scrape-able way is needed for the future of AI, but there should really be some way for the teachers of AI to get compensated for their efforts in improving it. If Stack Overflow trains a model on contributor data it should not call the contributors community, but investors with proper compensation.
It sounds like he’s suggesting that AI will open up a huge new pool of amateur developers, and those developers will need a community to turn to to know how to leverage AI.
But that kinda ignores the impact that AI will have on the concept of the SO community, or whether they’ll even be a need for it.
I honestly think his take is a misread of what kind of “community” SO has. My use of SO has always been 99% functional and borderline mercenary. I’m not getting to know anyone, I’m not building relationships, I just need a question answered. There’s nothing sticky about this community other than it being a good place to get those questions answered. As soon as AI can do that, I’ll never return to that “community”, and I’ll miss it as much as I miss Yahoo Answers.
I'm getting the point of the post, but confused to what it means for community/public SO.
Is the essence really: "Please contribute manual qualitative solutions to public SO, so we can use it to train GenAI for our enterprise customers", or have I misread?
I think this non-announcement is the preamble to trying to change the licensing of SO content in such a way that other organizations can't use it as training data, but SO can. Obviously such a move would outrage the SO userbase so they're flailing about trying to spin it as something to do with 'community'.
142 comments
[ 3.4 ms ] story [ 202 ms ] thread> This question has received 11 upvotes and 32 downvotes.
https://meta.stackexchange.com/search?q=score%3A-2000..-500+...
https://meta.stackoverflow.com/search?q=score%3A-2000..-500+...
The range is tinker able.
https://meta.stackexchange.com/help/privileges/established-u...
You can also get a lot of vote information in the data explorer https://data.stackexchange.com
I don't think most of the replies on SO mention this directly, they mostly focus on how this will impact the site's quality negatively. But really you are correct that people don't like being used as training data and imo the "arguments" everyone are making against AI disadvantages on SO, on art sites, on wherever else are pretty much "excuses". Quoted because even if they are correct they are just excuses for the core much more ethical issue - humans don't want to be replaced. Being replaced by a machine is even shittier. Being replaced by a machine that exploited your own hard work against you even more so.
In a real world example if my boss hired a junior whom I must mentor with the sole purpose of me being fired and replaced so he can pay him less. Well that would be considered very unethical even though it does happen and many workers do put up with it. Yet the current AI movement is that on an unprecedented scale and to the people being replaced it must feel like a massive bitch slap in the face.
To me it seems like an extreme expression of power and it's bizzare how many people are fine with it.
The name is Prashanth Chandrasekar or am I missing something?
This is a bad heuristic, strengthening your priors based solely off rejection.
Nobody questions LLMs can write code. The question is whether Stack Overflow has a place in that future. The community’s rejection, and CEO’s ham fistedness, suggest the data he has are the data he’s got. Which makes him uncompetitive vis-à-vis e.g. GitHub or any IDE.
Amen.
Most people actually like stability and no surprises, when doing their habits, like drinking coffee for example. It should not suddenly turn to wine, that would freak them out.
And books and movies should follow a familiar pattern, or it will be perceived as disturbing. Slight variations are welcome, but nothing too drastic. So I doubt people will be truly shocked.
Your 8 year olds cat drawing has been subsumed for 25c. Your essay on Philip K. Dick has been subsumed for .32c.
SO jobs should have been an absolute gold mine, instead he cancelled it.
Makes no sense to me. If you can’t make a ton of money on job ads on Stack Overflow then you’re not trying.
It’s true that SO was doing SO jobs wrong, but it should have been fixed, not cancelled.
job websites I can think of off the top of my:
hired
indeed
linkedin
ziprecruiter
google shows jobs
websites google recommends:
monster
glassdoor
simplyhired
yes none of these are customized like stackoverflow but, don't you think that competition is stiff enough?
Indeed can't find me devs that don't know they want to leave their current job yet.
Every business is recruiting and willing to pay hundreds or thousands to fill each spot.
Every single site you mentioned is filled to the brim with low quality low effort Contract to Hire recruitment spam.
StackOverflow could have been the way to get real tech jobs, but they squandered it.
If you can keep the quality high for a small-medium group there is money to be made, I imagine (i'm trying to start a job board myself for a tiny group). But just look at anything Pieter Levels has built (remoteok.io) dude has a huge monthly MRR. granted hes a solo founder and SO probably employs many. Also look at Dice.com. They allegedly bring in near 150Million/year! To have a targeted audience in the palm of you hands and throw it out the window is just bad business. SO is really just throwing away money.
The consensus at the time was that end to end wise (company hires and is happy with candidate) - you could just remove every integration leaving just stack overflow jobs and it'd be fine.
It's beyond me why they cancelled it, so stupid from my PoV.
SO should be ran by a foundation funded by tech companies enlightened enough to realize the massive productivity gains it produces (with SE as a side goodwill project). In this era, this counts as wishful thinking. Instead we got a 1.8B sale, that's not pocket change, now the profit must flow.
This is what Cory Doctorow calls enshittification. https://pluralistic.net/2023/01/21/potemkin-ai/
Tech is no more enlightened than the robber barons were 100 years ago.
There’s only three real options for platforms like SO:
- Mozilla
- Wikimedia Foundation
- Internet Archive
[1] https://blog.mozilla.org/en/mozilla/we-need-more-than-deplat...
[2] https://unherd.com/thepost/wikipedia-co-founder-i-no-longer-...
> Commit to meaningful transparency of platform algorithms so we know how and what content is being amplified, to whom, and the associated impact.
Can't really argue with that, even though I also frown at the tone of the test of the post.
But it is possible for tech companies, even ones we generally think of as evil, to collaborate (entirely out of self interest!) on mutually beneficial advancements.
Don't mistake "it would be wise for tech companies to support such a thing because they would get ROI" with "any company who supported such a thing should automatically be declared noble and perfect and of unimpeachable character forever"
In 2015 Prashanth Chandrasekar became Vice President and GM of the company -- the leader in all but name. In 2016 Apollo Global Management bought the company and took it private. The rest, as they say, is history. Sad history. And yes, he is 100% to blame. Or, from his and Apollo's perspective, to praise. Gutting a company this big brings in a lot of dough.
It took him two years to sell Stackoverflow. It's a harder sell, after all, than Rackspace.
We trialed a private SO instance at work, and discoverability was an issue (now we have to search slack engineering channels, check notion, and check SO).
If I could quickly ask, in for example slack, a question in natural language and have it query our private SO, slack, notion etc, I would be pretty game (providing this data remains private).
Yes, StackOverflow is not for beginners at all. It's for fairly narrow technical questions.
Maybe AI may assist in asking a good question or breaking down a big question into narrow on-topics ones.
All it can provide is how the answer would sound or look like.
Be prepared for an absolute avalanche of bugs and security holes no one alive would've made. I make my living from debugging so I welcome the new job security but it's a massive net negative for society, no question about that.
This includes providing correct answers, because by definition, a correct answer sounds exactly like it would sound like.
The answer still has zero information value, even if it is accidentally the correct one.
Next, it tried to use AlpineJS to create tooltips with NextJS which is really not supported and couldn't fix the resulting bugs.
For sure, it's better than nothing, but you've got to be at least somewhat competent before you start copy-pasting code from it blindly.
But that isn't the only task available for fancy auto-complete. Eg. The LLM could help novices make better questions, or include more/less context in an answer, with the person still being the arbiter of truth.
I am sure there are older internet examples as well
Edit: I love your suggestion as well, any UI UX that helps surface existing relevant content to would be question askers is golden and very important
Great suggestion!
I don’t know if what he’s suggesting here makes sense (I tend to think no), but I’m automatically a little skeptical of the typical response to a serious threat: “no way, y’all, this is actually good for us!”
Imagine google coming out right now and saying the future of AI is search ads.
But that kinda ignores the impact that AI will have on the concept of the SO community, or whether they’ll even be a need for it.
I honestly think his take is a misread of what kind of “community” SO has. My use of SO has always been 99% functional and borderline mercenary. I’m not getting to know anyone, I’m not building relationships, I just need a question answered. There’s nothing sticky about this community other than it being a good place to get those questions answered. As soon as AI can do that, I’ll never return to that “community”, and I’ll miss it as much as I miss Yahoo Answers.
Is the essence really: "Please contribute manual qualitative solutions to public SO, so we can use it to train GenAI for our enterprise customers", or have I misread?