The Fall of Stack Overflow

You trust LLM responses? I feel they are not worth the paper they are written on

I recall that about ten years ago there were reports that edits to Wikipedia were slowing precipitously. People wondered if this meant the imminent death of Wikipedia. With the benefit of hindsight, it seems more plausible that Wikipedia had merely shifted from its adolescent phase to its mature phase, and that all the low-hanging fruit had simply been harvested. If we look at the stats, Wikipedia's edit velocity maxed out in early 2007 (37 days per 10M edits), then fell to half that in late 2014 (73 days per 10M edits), and has actually increased a fair bit since then: https://en.wikipedia.org/wiki/Wikipedia:Time_Between_Edits

It's entirely possible that the same is happening to SO.

I wonder if this might be partly because they have not made an effort to make their software be more intelligent about surfacing answers to questions as well as evaluating answers. I would think that, at some point, it isn't necessarily the fault of the question authors or those responding.

One of the things I say with some frequency is that all failures are engineering problems. Blaming the users/customers, in the end, fails to take advantage of an opportunity to improve the product, learn from mistakes and improve process.

You don't necessarily need to trust them for them to be useful.

I can often evaluate a Copilot autocompletion (check that code looks right at first glance, check that it compiles, hover over method+type signatures to see their docs, run the code) in less time than it would take me to find+read a Stack Overflow answer.

Interesting that this seems to have started around Spring 2021 when posts + votes started tailing off, followed by traffic starting to decline around spring of 2022.

I can think of a few theories that I don't think hold water:

1. The rise of ChatGPT to answer many questions that StackOverflow would previously have been used for. This seems unlikely, since the timing doesn't really work out.

2. The perennial complaints about StackOverflow's culture of closing everything as duplicates or offtopic. This seems unlikely as well, since those complaints have been common for a decade or more.

3. The prevalence of SEO-optimized scrape sites - the ones that pop up with a "blog post" merely reposting a stack overflow question + answer in a different font. I've seen these for a while, and anecdotally they feel more common that they used to, but I couldn't give any real timeline for that vague feeling.

4. StackOverflow internal politics? I've seen the occasional stack-overflow meta thread pop up periodically on HN or social media, but I don't recall anything earth-shattering recently.

5. Most questions have good answers now and there's less need for new ones. I'd have bought that answer 10 or so years ago when StackOverflow's pile of questions + answers reached maturity. I don't think it suddenly hit some sort of answer saturation point in 2021.

My guess is that it's a slow shift in the culture of the StackOverflow userbase:

- Being a top answerer confers some cachet and makes you more employable in some places

- People notice this and start looking for the most effective ways to become a top answerer

- The most effective way is fast, low-effort answers

- There's been a rise in such low-effort answers over the last 5 years or so

- As a result, the cachet of being an StackOverflow top answerer is a lot lower

- The really good, deep, technical answerers (as well as the mods) are leaving as that cachet goes away

- Post quality starts dropping around 2021 and views start declining as people react to that in 2022.