Comment by rerdavies | Hacker News Reader

Comment by rerdavies | Hacker News Reader

The biggest problem with Stack Overflow is that both the questions and the answers age poorly. A question that was asked and answered for Android 7 is no longer relevant for Android 14. As a result, many, if not most top answers are no longer relevant.

I find these days that the answers I'm looking for are increasingly answers that have snuck in during the brief period before questions are flagged as duplicates. For most original questions, the top ten posts are so old that they are no longer relevant.

StackOverflow would be so much better if questions that were flagged as duplicates were not closed for further answers. Flagging a question as a duplicate should be an advisory, not a death knell.

There's also a huge problem with questions that are not duplicates being flagged as duplicates, because there's no penalty for editors being wrong, and no easy way to appeal that decision. Determining whether an answer is a really duplicate often involves close reading of both the original and the new question -- something that seems not to happen in the majority of cases. The problem is compounded because editors have a direct incentive to flag non-duplicates as duplicates. And so, the quality of answers continues to decline.

As a concrete example, my StackOverflow account has accumulated about 5,000 points for an answer I wrote for Android 7 that describes diagnostics that were changed in Android 8. Nevertheless, it continues to accumulate about 100 points a month even today. I can't be bothered to fix it, because it would involve significant research. The contents of the diagnostic results aren't documented. I had to go to Android source to figure out what was actually being reported. And it's not a topic that particularly concerns me at present. If someone were to go to the trouble of analyzing Android 14 sources to update my post, I doubt they would ever be able to overcome the very significant number of upvotes my answer over the last 8 years.

This explains why the only currently functioning answer to many questions I have often has 0 votes and is half way down the page.

> A question that was asked and answered for Android 7 is no longer relevant for Android 14.

SO power (top 0.07%) user here. Here's how I handle that:

- If adding a new answer to an old question, add a headline at the top of your new answer explaining that it's new for Android 8, post 2017, ES6, whatever else.

Here's a JS vs jQuery answer. 'Trending' shows the 2014 (new) plain-JS answer before the older (2011) jQuery answer:

https://stackoverflow.com/questions/2276463/how-can-i-get-fo...

- Generally this should be enough to handle fixing things, but if you still use SO and are aware you have an old answer, and don't want to update it, add a headline saying 'This is for Android 7, pre 2017, ES5 and older' etc.

So true. I feel like I used StackOverflow so much when I started programming, and now it’s rarely even the most useful result on Google for the question I’m trying to answer. I struggle to imagine a scenario now where I’d want to rely on fossilized SO instead of GPT cross-referenced with official documentation.

> fossilized

Another choice of vocab: ossified

I think it's easy to forget that as we get better at programming, we tend to get better at either avoiding problems or finding our own answers in which case SO is also less useful because we've learned to fish.

Yep. A decade into my career, the problems I tend to face are often ones that aren’t covered on SO, and probably wouldn’t be “appropriate” for SO, either. Something something ‘opinion-based’.

I had this realization a few years ago compared to other peoples' general search strategies.

An MIT professor on OCW (intro to machine learning) said something that has stuck with me - in paraphrase: "In order to learn something we must give it a name."

Now I just intuit which search terms work and can scan the results to find leads. A lot of the results and examples take coaxing to work, but sometimes you find that diamond that perfectly addresses your problem.

Which professor?

I really wish I could point you to a timestamp. I think it was this series.

https://youtu.be/TjZBTDzGeGg?si=jNbm7K3e--buioPp

They do a sort options:

Highest score (default)

Trending (recent votes count more)

Date modified (newest first)

Date created (oldest first)

I've found the version issue to be significant on SO and in documentation in general when accessed via a search engine. Sometimes, it's "does this solution still work at all?". Other times, it's "is there a better or preferred way to solve this in later versions?"

For SO, seems a "version applicability" option could go a long way to solving this.

SO has basically been obsoleted by LLMs at this point. Why even visit the site? They are clearly not interested in improving the UX for either questioners or responders, or the 'Closed as duplicate of a question that somebody answered poorly 8 years ago' policy would have been changed or scrapped long ago.

One obvious question, of course, is how the LLMs are going to keep themselves up to date. It's not as if Stack Overflow is doing that, given the points raised by the parent. It seems that we're headed for stasis unless a new developer Q&A paradigm emerges.

LLM are trained on stack overflow.

Um, yes? That's kind of my point. LLMs trained on SO answers from a time before SO sucked as badly as it does now. Where will next-gen LLMs get their training, without a good open Q&A site to leech from?

LLMs are still being trained on StackOverflow. Even very recent Stack content can improve LLMs

And that's why I use it so much for getting my way around the Linux shell :)

Reply on news.ycombinator.com