| So here's the fun part. I'm all over this thread explaining how Stack Overflow is supposed to work, and how almost everyone who comes to it has broad misconceptions, and about how we're very deliberately gatekeeping in a specific way for a specific purpose, and why I think that purpose is legitimate. I've also expressed quite a bit of AI skepticism on HN before. And I absolutely agree with you - assuming that by "use Stack Overflow" you mean "post a new question". Most questions are fundamentally simple, even if they're tedious. If you're implementing something "complex", usually it really just involves following a straightforward series of steps, and it just takes a little bit of experience to see what the steps are. That breakdown is probably not useful to anyone else, and is hard to turn into a searchable artifact for others. If you're tearing out your hair troubleshooting something, most of the time one of the "standard" fixes really will work, and you just need to be talked through them. If you like reading professional/academic sounding prose extensively trained to be inoffensive (even if it isn't always to the point), you have much better odds now with an LLM than with a random person who happens to be an expert on whatever it is you're trying to do. If you're trying to figure out a thorny problem and it would benefit you to have a back-and-forth communication - or ask multiple questions on a theme and get all the responses from the same place - Stack Overflow can't do that for you. In fact, traditional forums can't, either, except accidentally (because they're too small to expose you to more voices). But the Stack Exchange Q&A model is explicitly about preventing back-and-forth communication, because it degrades the experience for third parties. If you take downvotes personally (they aren't meant personally) or expect that your questions should be answered simply because you ask them, then of course you should prefer a system that doesn't rate what you say (unless you explicitly ask) and is directly tasked with responding (even if you prompt with complete nonsense). And if you want a system that tries to adapt code to your personal circumstances and minimize the editing you need to do, then you obviously shouldn't use one that specifically tries to show demonstration code relevant to everyone with the same problem. ---- But the LLM is trained in part on Stack Overflow content. And if everyone using Stack Overflow used it as intended (and as took years to properly figure out, and is taking much longer to communicate), the LLM would be trained on much better Stack Overflow content. And would also not need the LLM a lot of the time, because a traditional search engine would find you high-quality Stack Overflow content directly. (Actually, despite everything that's gone wrong, traditional search engines still do a reasonable job a lot of the time. Granted, it's extremely frustrating when they fail. I know because I've experienced this failure when trying to find a good target to close an obvious duplicate.) And sometimes the LLM will be wrong; and it will be absolutely abysmal at introspecting whether it's wrong. |