Stack Overflow is no longer relevant. Today, you can just ask Gemini, Claude, or ChatGPT instead, and you don’t have to deal with the usual condescension.
I can see where you might get that sentiment, but where do you plan to go when new tech rolls around, the docs don't cut it and your LLM of choice hallucinates APIs that don't exist? This was always Stackoverflow's bread and butter, and people who only use it as noob search tend to miss that fact. SO can be a tough crowd, yes, but mostly it's people who didn't read the rules before posting who get burnt. That aside, it still has a very high concentration of experts that you'll struggle to find anywhere else.
>I can see where you might get that sentiment, but where do you plan to go when new tech rolls around, the docs don't cut it and your LLM of choice hallucinates APIs that don't exist?
Not Stackoverflow, because all my questions are either ignored or closed, even when extremely detailed and unique.
Well my last attempt turned out to be a vendor issue. It was officially acknowledged by a vendor support rep. It had to do with their unique method of populating a whitelist with DNS entries, and weird response chain to blocks in that state.
I read through the backlog, ensured I had completely exhausted every other avenue (short of 3 more weeks of yelling at one of the involved vendors for information about how they were performing their whitelisting) and had available captures for network and application, full reproducibility steps and details of everything that had been interrogated. I even remember linking similar issues, and explaining how I ruled out their causes.
In fact I am still relatively confident that the way mobile browsers were responding to the bug, probably constitutes a bug in itself but honestly, cant be assed to pursue it.
When enough time has passed I will probably produce a blog about the issue, so it can be digested by the next iteration of the troubleshooting machine. But I really dont feel compelled to provide further data to the stackoverflow community directly considering their complete lack of response. Even closing it as a duplicate of an existing issue would have been helpful, but it wasn't a duplicate so it was just ignored.
Unfortunately the answer to that is the Discord server of whatever technology I'm working with. Communities are now separated each in their silo on Discord, far away from the public internet, where nothing can be indexed.
Yet. By the time stackoverlow shuts down, AIs will be powerful enough to take data from docs or just from the source code alone. I mean the new version of opus is pretty good at understanding my front end source code. I think that should be the goal of AIs (that they are so advanced they don’t need to read code examples from a third party website like stackoverflow)
This is a pretty narrow view. Most APIs out there are not source available, frontend libraries are the odd one out in that regard. Likewise for docs, it doesn't matter if the LLM can read the docs or not if the docs literally don't include the things you need to figure out. I suspect this is a generational divide, people who grew up with SO can't imagine what life was like before it, but right now we're on a straight course back to isolationist communities, if we aren't already there.
While LLMs may have used Stack Overflow data to get their start, I think it's reasonable to assume that this source of training data will no longer continue to be useful.
Therefore, as both a data source and a QA website, Stack Overflow has lost its relevance.
If an LLM can read the source
of the library you’re trying to use - or examples of others using the library in GitHub, or official documentation - then there is less of a need for a fellow SOer to put the pieces together to debug issues and answer questions.
My niche questions are never answered correctly by AI. I'm led down false rabbit holes. Stack overflow still provides much better answers for me overall.
Growing grains is no longer relevant. You can just walk into any supermarket and purchase packaged cereals, breads, and cakes, and you don’t have to deal with operating a tractor, cultivating soil, or sowing seeds.
SO is probably a very significant factor in the success of LLMs but it's decline will not affect LLM development. LLMs will simply be trained on the conversations people are having with them.
Unlikely to happen with conversations, yes, but very likely to happen implicitly as part of work with coding agents. Successful agent trajectories will be used to learn what works and what doesn't for a given API.
