The decline began before ChatGPT's November 2022 launch. So it would still decline even without the availability of LLMs. For me personally SO has mostly lost it's value because of the bueraucratic and arrogant ivory-tower mentality of the admins.
> "cannibal's dilemma"
Looking at services like Perplexity, I don't think it is/was essential to use SO data for training, since the main point of Perplexity is to use sufficiently intelligent agents (based on LLMs) to search and infer from many sources. The specific knowledge the agent received from the training data is not only static (i.e. not current), but neither essential, as long as the agent has sufficient general knowledge to solve the research request. It's like real experts; the knowlege we were trained on at university is only a small subset of the world and usually not very current; but we have learned how to do research and to solve problems. I'm much more efficient with Perplexity than I ever was with SO, and no longer silly discussions.
SO was always explicitly intended to be an "ivory tower", or at least to produce one as a permanent, useful artifact. The people you call "admins" are not anything of the sort; we have a few dozen actual moderators, and then a hundred thousand or so people entitled to cast close votes, and many more entitled to edit posts unilaterally. The role "administrator" does not exist at Stack Exchange. There are staff who actually work for Stack Exchange, Inc., but in large part they are enemies of the community on meta who discuss policy and try to organize curation.
> I'm much more efficient with Perplexity than I ever was with SO, and no longer silly discussions.
SO is explicitly not for discussion of any kind. That's why there's a meta site, so that even questions about SO itself don't have to distract from the main Q&A flow.
If you wanted to talk to some"one" (whether human or AI) back and forth to solve a personal problem, then SO was never intended to fill that purpose, and in fact explicitly designed to avoid that mode of discourse (even though the original design created some perverse incentives that were never addressed). The whole point of it is to not be a discussion forum - so that when you find a Q&A later, you don't have to read through a "thread" to find the question, filter out the arguing, make sure you found something relevant, and get an answer. So that when you want to be "helpful", you can explain a basic topic once and refer people back to it later; and so that you can directly answer questions instead of trying to figure out what people even want to know in the first place.
If you were on SO because you wanted to describe your current situation and have someone help you fix the code, you were always in the wrong place, and it was your responsibility to understand that.
If the site "lost value" for you in that regard, it's only because people were trying to provide value that was never in the site's mandate - in a way that directly harms the site's actual goals.
Codidact (https://software.codidact.com), for those who actually care about the Q&A site model (although we're designed explicitly to allow for opening separate "sections" for things that don't work like Q&A). That is, if you're posting a question in order to frame useful, searchable information, and not simply to get help with a project.
Disclosure: I am a moderator there (actual moderator, not just someone entitled to edit posts - which we still do as a community, with the same kind of Creative Commons licensing arrangement).
I bet Google are kicking themselves for killing Google Code.
https://code.google.com/archive/about
> But alas, in 2015 Google bid farewell to the service
