Live data from Hacker News

Stack Overflow is almost dead

blog.pragmaticengineer.com

81–90 of 285 posts

Re: Stack Overflow is almost dead

#81
My day job includes moderation. I think I am more empathetic to the issues that moderators deal with on a day to day basis given that they are underpaid, under-appreciated, and overworked.

It's very difficult to scale a community to be both welcoming and productive. New users don't have the same context as existing ones. You find that norms and manners aren't transferred from one group to the next. So although that I noticed that SO started getting more strict from 2014 onward, I wouldn't know immediately what to do about the content quality issue.

My take is that, like most things, the medium of the old will be appreciated the way it wasn't in 2014. As the Brian Eno quote goes: "Whatever you now find weird, ugly, uncomfortable and nasty about a new medium will surely become its signature." People will yearn for the human forums the same way they did years past when people tire of the LLM slop. (If they do.)

Re: Stack Overflow is almost dead

#84
This will have interesting implications for the LLMs as well, since SO is a wealth of training data. In my experience, LLMs are pretty useless when it comes to helping me with newer, faster-evolving, experimental tools and libraries, which is not surprising. But, if the SO community really atrophies to the point that a lot fewer people are bothering to answer questions, there won't be another centralized resource for answers. Perhaps that just means balkanized communities like random Slack channels will fill the gap, but those aren't search-indexed and I'd bet getting them all into training corpi won't be as easy either.

Maybe the future involves LLMs asking questions on something like SO when it routinely fumbles a particular topic. People could get paid to answer them and provide more training data. Who knows at this point

Re: Stack Overflow is almost dead

#85
How ironic. "AI" feeds off structured knowledge, artistic creations and otherwise any human production to generate its output. As a consequence of its widespread adoption, people start to lean even more towards consuming rather than producing, a tendency which was already increasing before the advent of LLMs and modern machine-learning. This, in turn, leaves "AI" implementations with no new human content to feed off of. Now what? The whole process folds onto itself. Are we entering the dark ages of cultural (in the widest sense of the word) production? Not that I don't think that we're already there, in any case, but for other, somewhat related causes...

Re: Stack Overflow is almost dead

#86

My heart goes to the stack overflow community which has always been very kind and helpful, essentially working for free. As a self-taught developer since the age of 8, I literally grew up learning how to code through SO, asking hundreds of questions and answering many more. So many bugs that would take 2-3 days to fix would eventually find their answer through it. But now ChatGPT does that in minutes… so it’s for the…

> But now ChatGPT does that in minutes But it's trained on stackoverflow data? What happens in a few years when the data gets more and more outdated? Where will it get its knowledge then?

Why does any LLM need new information to do fundamentally the same thing?

And what makes the data outdated? New code? It can train on that. That, or there is simply nothing new to learn, just new ways to express the same thing.

Re: Stack Overflow is almost dead

#87
Just how is 25K new questions a month dead? Even if it gradually asymptomates to just 1K, the answers to them are enough to continue serving as a critical base layer of high-quality training data for LLMs.

Let's say for the sake of argument that 95% of humanity perished. Is humanity then dead? It isn't.

Re: Stack Overflow is almost dead

#88
post #14
post #9

I find it interesting that the current StackOverflow moderators tend to say "in the past we used to accept too many questions but it was never the goal, so now we are doing it as it was meant to be". Sure, but in the past, StackOverflow was growing, and now it's dying. Maybe something was better before, when "it was not done correctly"?

AFAICR they've always said these lines about now is about better moderation from the slop. The reality is that the rule of thumb for that moderation was already out of date with advances that preceeded LLMs.. Even with the beginnings of computer aided flows we didn't need to alienate most to get the best content and develop the few. Content can be triaged from someone who may be human to others who may be human and m…

> Even with the beginnings of computer aided flows we didn't need to alienate most to get the best content and develop the few.

The large majority of new questions from new accounts are from people who are clearly there only to solve a personal problem, who show no interest in considering the value of their question to third parties, and rarely put any effort into attempting to even diagnose or specify a problem.

Even after it became possible for most of these people to get an instant answer from an LLM. Which is actively preferable from the standpoint of Stack Overflow curators. Before LLMs, the point was for them to use a search engine to find an existing question that lets them figure out the problem. But for the Q&A to help such users, they need to apply at least basic problem-solving and debugging skills. (It is explicitly out of scope for the Stack Overflow community to do that for others; and attempting to do this in an answer actively degrades the site for everyone else.) If an LLM can fill in some hypotheses for those users to test, then the LLM is doing what it's best at, and Stack Overflow is doing what it's best at.

Stack Overflow is not there to troubleshoot or debug anything for you, nor to reason about a multi-step problem and break it down into its natural logical steps. It's there to give a direct, objective answer to how to do each individual step, and to explain why the specific point of failure in a failing program fails, after you have identified it and made the problem reproducible.

So yes, we absolutely do need to "alienate most", because "most" are there for a reason that has absolutely nothing to do with getting the best content.

Re: Stack Overflow is almost dead

#89

> 2014: questions started to decline, which was also when Stack Overflow significantly improved moderator efficiency. From then, questions were closed faster, many more were closed, and “low quality” questions were removed more efficiently. This tallies with my memory of feeling that site moderators had gone on a power trip by closing legitimate questions. I stopped asking questions around this time because the site…

Yes I felt the same around then. It seemed like stack overflow was sending the message to not ask questions anymore. It was really weird.

Re: Stack Overflow is almost dead

#90

My heart goes to the stack overflow community which has always been very kind and helpful, essentially working for free. As a self-taught developer since the age of 8, I literally grew up learning how to code through SO, asking hundreds of questions and answering many more. So many bugs that would take 2-3 days to fix would eventually find their answer through it. But now ChatGPT does that in minutes… so it’s for the…

I just hope that we can continue to find sources of high quality training data like SO. If people don't publish their mutual learnings somewhere then there's no data to train on.
Post reply on HN