We might have been slower to abandon StackOverflow if it wasn't a toxic hellhole

Well put. It's maddening that people still don't get it 17 and a half years later. (It's interesting that Spolsky is giving the talk, since to my understanding Atwood was considerably more hard-line about it, and was always the one blogging about the high concept.)

I mostly chalk it up to UI affordances. The most obvious one: the site constantly presents an "Ask Question" button; it gives you a form to type in a question; people come to the site because they have a question, and it goes live to a general audience[1] as soon as it's posted. No amount of emphasis on search is ever going to override that.

Less obvious but much more important is that the community can't actually put information about community norms in front of new users, except by scolding them for mistakes. No matter how polite you are about giving people links to the tour[2] or to policies[3],

Then of course, they wanted the site to actually grow at the start, so we got that terribly conceived reputation system best described as Goodhart's law incarnate[4]. And it was far too successful at that early growth, such that if anyone actually understood the idea properly at the start, they were overwhelmed by new users (including the experts answering questions) and had no chance to instill a site culture. It took until 2012 or so until a significant chunk of the experts were getting frustrated with... all the same things they were historically frustrated with on actual forums; then we got the "What Stack Overflow is Not" incident[5]. A lot of the frustration was misdirected except for a general annoyance at certain stereotypes of typical users. It took until at least 2014, from my assessment of the old meta posts, for a real consensus to start emerging about what makes a good question, and even then there was a lot of confusion[6].

Newer sites like Codidact[7] have a chance to learn from this mess, establishing ideas about what good questions look like, and about site scope, from the start[8].

1. Notwithstanding more recent efforts, like the Staging Ground and now a new "question type" feature (https://meta.stackoverflow.com/questions/435293) which seems to have been recently rolled back in preparation for something bigger (https://meta.stackoverflow.com/questions/437856), and the various attempts to force AI into the process, etc.

2. https://stackoverflow.com/tour

3. Especially things like 'Under what circumstances may I add "urgent" or other similar phrases to my question, in order to obtain faster answers?' (https://meta.stackoverflow.com/questions/326569) and 'Why is "Can someone help me?" not a useful question?' (https://meta.stackoverflow.com/questions/284236). See also https://news.ycombinator.com/item?id=46485817 .

4. https://meta.stackexchange.com/questions/387356/the-stack-ex... ; the anchor is for my own answer but please scroll around and read other points of view.

5. See https://meta.stackexchange.com/questions/137795. Back then I was actively using the site but not active on meta; I pretty well gave up in 2015 for largely unrelated (personal) reasons, then came back in mid 2019, coincidentally shortly before the Monica situation[9].

6. In particular, see 'How much research effort is expected of Stack Overflow users?' (https://meta.stackoverflow.com/questions/261592), originally authored 2013, and especially compare the original answers to newer ones. Notably there were also quite a few deleted answers on this one, for those of you with the reputation to view them. Also see 'How do I ask and answer homework questions?' (https://meta.stackoverflow.com/questions/334822) which largely misses the point: it's not so much about the ethics of someone cheating on homework, but about the question fitting the site model.

7. https://codidact.com , with subdomains for various topics. Notably, "programming" as a topic is not privileged; unlike how the Stack Exchange network started with Stack Overflow which still dominates everything else put together, software.codidact.com is just another section of the site. Full disclosure: I am a moderator for that section.

8. See for example https://software.codidact.com/posts/285035/289176#answer-289... ; https://software.codidact.com/posts/291064 ; https://software.codidact.com/posts/284979 ; https://software.codidact.com/posts/292960 ; https://software.codidact.com/posts/294610 ; https://meta.codidact.com/posts/289910 ; https://meta.codidact.com/posts/290028 ; https://meta.codidact.com/posts/291121/291156#answer-291156 ; https://meta.codidact.com/posts/289687 ; https://meta.codidact.com/posts/289951 ; https://meta.codidact.com/posts/284169. Yes, this is a carefully hand-picked list. I have a fairly clear mental image of one more but was somehow unable to search for it.

9. See https://meta.stackexchange.com/questions/333965 and many others. It's a deep rabbit hole. It's also the triggering incident leading to the creation of Codidact[7].

The duplicate flagging is definitely the most common bad moderation I saw on SO, for sure. Good point.

You are now often debugging problems directly in an LLM context window, providing back console output, error logs, etc. there is an absolute mountain of data to mine from developer chat logs.

This is not my experience. So many comments are flagged and removed. It’s just popularity.

I think it’s fine they are hidden by default. But unt we can see all removed comments we can’t understand the debate.

1. They overwhelmingly are not moderators, and they are not doing moderation by closing questions. This is curation, and duplicate closures overwhelmingly are done by subject-matter experts: users with a gold badge in one or more of the tags originally applied to the question. The requirement for such a badge is based on answering questions:

> Earn at least 1000 total score for at least 200 non-community wiki answers in the $TAG tag. These users can single-handedly mark $TAG questions as duplicates and reopen them as needed.

So these are definitely not people averse to the idea of answering questions.

2. I can guarantee you that the overwhelming majority of these cases are not people trying to be "mean". Users are actively incentivized against closing duplicates, which has historically led to nowhere near enough duplicate questions being recognized and closed (although there have been many proposals to fix this). Dupe-hammering questions "to be mean" is considered abusive, and suspicion of it is grounds to go to the meta site and discuss the matter.

No, people close these questions because they genuinely believe the question is a duplicate, and genuinely believe they improve the site with this closure. It's important to understand that: a) people who ask a question are not entitled to a personalized answer; b) leaving duplicate questions open actively harms the site by allowing answers to get spread around, making it harder for the next person to find all the good ones; c) the Stack Overflow conception of duplication is not based on just what the OP understands or finds useful, but on what everyone else afterward will find useful.

For example, there are over a thousand duplicate links to https://stackoverflow.com/questions/45621722 , most of which is from my own effort — spending many consecutive days closing dozens of questions a day (and/or redirecting duplicate closures so that everything could point at a "canonical"). Yes, that's a question about how to indent Python code properly. I identified candidates for this from a search query and carefully reviewed each one, verifying the issue and sending other duplicates to more specific canonicals in many cases (such as https://stackoverflow.com/questions/10239668). And put considerable effort into improvements to questions and existing answers, writing my own answer, and adding links and guidelines for other curators so that they can choose more appropriate duplicate targets in some cases. I also looked at a wider search that probably had a fairly high false positive rate, but implies that there could be thousands more that I missed.

3. When your question is closed as a duplicate, you immediately get a link to an answer. You don't even need to wait for someone to write it! It's someone saying "here, I was able to find it for you, thanks perhaps to my familiarity with other people asking it".

4. Stack Overflow users really do "try to find what's different about" the question. It just... doesn't actually matter in a large majority of cases. "I need to do X with a tuple, not a list" — well, you do it the same way. "I need to Y the Xs" — well, it seems like you understand how to Y an X and the real problem is with finding the Xs; here's the existing Q&A about finding Xs; you shouldn't need someone else to explain how to feed that into your Y-the-things loop, or if you do, we can probably find a separate duplicate for that. Things like that happen constantly.

Sometimes a question shows up with multiple duplicates. This almost always falls into two patterns: the user is really asking multiple separate things (due to failing to try to break up a problem into logical steps) and each one is a duplicate; or the question is constantly asked but nobody knows a good version of the question, and gives multiple links to previous attempts out of frustration with constantly seeing it. (The latter is bad; someone is supposed to write the good version and send everything else there. But that typically requires behind-the-scenes coordination. Better would be if the first bad attempt got fixed, but you know.)

5. Closing a question is emphatically not about telling people to go away. The intended message (unless the question is off topic or the OP just made a typo or had a brainfart) is "please stay and fix this". However, it's perfectly reasonable that an explicit attempt to catalog and organize useful information treats redundant indices by pointing them at the same target rather than copies of the target. And questions are indices in the Q&A model.