| > It's kind of related because AI tools are for better or worse going to eat up a bunch of these question/answer responses from tools like Stack Overflow for developers, particularly for trivial things. I don't see the connection between 58 people being laid off and a rep farming howto. If the parent commenter is trying to initiate a discussion about whether AI is a contributing factor in this round of layoffs at Stack Overflow, that's relevant and welcome, but I don't see any indication of that from the post. Stack Overflow and LLMs have different use cases, with some overlap. They're complimentary resources. SO is not a free code-writing service, although it's often been mistaken for and abused as such. "How do I create a list of HTML Bootstrap cards for my whatever app", would be closed as off-topic. It's basically a work request for a freelancer or an LLM. The goal of SO is to curate a repository of programming knowledge based on focused, specific, well-researched technical questions. It's served me very well in this purpose over the years and continues to post-LLM boom. Many questions deal with up-to-the-date issues that are beyond LLM training sets, and crowd-vetted answers generally don't hallucinate. I'm happy to see LLMs take care of the trivial cases you mention. They're good for many drudge tasks. Typos, syntax errors and misunderstandings that would be resolved by a quick glance at the docs or an LLM are not really a good fit for SO, unless they're common. Prompting an AI should be among the expected prerequisite attempts before asking SO. A recent personal example of SO at its best was a question[1] based on a regression bug caused by a couple-day-old deploy to the Puppeteer browser automation library, a tag I answer questions in daily. I reproduced the problem, forked the repo, fixed the issue, opened a PR and the patch was deployed to NPM the next morning. Given the same input, LLMs will pretend like they know what they're talking about and give you the run-around endlessly, providing random, plausible-looking nonsense for as long as you have patience to put up with it. Only humans can handle this type of problem at the moment. And this isn't an isolated example. > hearing that this is possible stings because it really cuts through the veil. Cheating with LLMs is not possible currently. These answers stick out like a sore thumb. It's just more work for the community to clean up the LLM garbage being dumped on us. I'm not sure what "veil" you're referring to--the reason LLMs are banned is because they don't meet the site's standards for quality and accuracy. If LLMs were objectively better at doing what SO does across the board, SO would be a ghost town by now. Some day, LLMs may well replace humans (fully, or to a large extent) at programming, but I don't think it'll happen as soon as commenters seem to imply, judging by the quality of LLM answers I've seen and my experience using ChatGPT. And if LLMs do replace humans, fine, life will go on, but that day hasn't arrived yet so it's premature to bury human programming Q&A in general or SO specifically. [1]: https://stackoverflow.com/questions/75292123/why-does-puppee... |
A flow in the past might have been Google “question content or error message or something” -> Stack Overflow links to browse through until something plausible shines through -> problem solved
Now it’s more like Chat-GPT (problem probably solved, or additional details to Chat-GPT) -> Google -> Stack Overflow
Even if tools like Chat-GPT were half as good as Stack Overflow they are vastly more efficient and since I know what I’m doing or looking for I can sniff out how plausible something is (no difference with Stack Overflow today).
I get that you have had a great experience on the site and truly it delivers value for a lot of people!
But I don’t see how such things keep the lights on at Stack Overflow.
The reason the OP is relevant is because even if you think you can identify the answers today it won’t be long before you can’t. It follows the principal that it’s easier to destroy stuff than build it.
If Stack Overflow wants to survive in any form it probably needs to Verify Human and eliminate any method in which someone can interact with the site (content-wise) except hands on keyboard.
Personally I’ve found the so-called human responses to be +- useful compared to Chat-GPT on any given topic I use it for and the only way I come to believe they’re human responses is because of the time stamp.
For example you say:
> …LLMs will pretend like they know what they're talking about and give you the run-around endlessly, providing random, plausible-looking nonsense for as long as you have patience to put up with it
I honestly think you can just plug in “contributors to most Stack Overflow questions” in to this sentence and it’s equally true.
You are focusing on the 1% of cases like in your example. I’m looking at most cases.