Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Large language models reduce public knowledge sharing on online Q&A platforms

academic.oup.com

Post reply on HN
Print view View original

341–350 of 366 posts

Re: Large language models reduce public knowledge sharing on online Q&A platforms

#341

Post by zahlman »

wokwokwok wrote: post #109
beeboobaa3 wrote: post #101

Earlier quoted context omitted.

I'm sorry but the funny thing is, the only people I've ever seen complain about SO are people who don't know how to search.

Everyone has a pet theory about what’s wrong with SO; but here’s the truth: Whatever they’re doing, it isn’t working. Blame mods. Blame AI. Blame askers… whatever man. That is a sinking ship. If you don’t see people complain about SO, it’s because they aren’t using it, not because they’re using the search. Pretty hard to argue at this point that the problem is with the users being too shit to use the platform. That’s…

>Whatever they’re doing, it isn’t working.

It's working just fine. The decline in the rate of new questions is generally seen as a good thing, as it's a sign of reaching a point where the low-hanging fruit has been properly picked and dealt with and new questions are only concerned with things that actually need to be updated because the surrounding world has changed (i.e., new versions of APIs and libraries).

>Pretty hard to argue at this point that the problem is with the users being too shit to use the platform.

On the contrary. Almost everyone who comes to the site seems to want to use it in a way that is fundamentally at odds with the site's intended purpose. The goal is to build a searchable repository of clear, focused, canonicalized questions - such that you can find them with a search engine, recognize that you've found the right question, understand the scope of the question, and directly see the best answers to the highest-quality version of that question. When people see a question submission form and treat it as they would the post submission form on a discussion forum, that actively pollutes said repository. It takes time away from subject-matter experts; it makes it harder for curators to find duplicates and identify the best versions thereof to canonicalize; and it increases the probability that the next person with the same problem, armed with a search engine, will find a dud.

Re: Large language models reduce public knowledge sharing on online Q&A platforms

#342

Post by zahlman »

Ferret7446 wrote: post #202
wokwokwok wrote: post #109

Earlier quoted context omitted.

Everyone has a pet theory about what’s wrong with SO; but here’s the truth: Whatever they’re doing, it isn’t working. Blame mods. Blame AI. Blame askers… whatever man. That is a sinking ship. If you don’t see people complain about SO, it’s because they aren’t using it, not because they’re using the search. Pretty hard to argue at this point that the problem is with the users being too shit to use the platform. That’s…

I submit that what SO is doing is working; it's just that SO is not what some people want it to be. SO is not a pure Q&A site. It is essentially a wiki where the contents are formatted as Q&As, and asking questions is merely a method to contribute toward this wiki. This is why, e.g., duplicates are aggressively culled.

>is not a pure Q&A site. It is essentially a wiki where the contents are formatted as Q&As

The thing is, the meta community of Stack Overflow - and of other similar sites like Codidact - generally understand "Q&A site" to mean the exact thing you describe.

The thing where you "ask a question"[0] and start of a chain of responses which ideally leads to you sorting out your problem, is what we call a discussion forum. The Q&A format is about so much more than labelling one post as a "question" and everything else as an "answer" or a "comment" and then organizing the "answers" in a certain way on the page.

[0]: which doesn't necessarily have a question mark or a question word in it, apparently, and which rambles without a clear point of focus, and which might be more of a generic request for help - see https://meta.stackoverflow.com/questions/284236.

Re: Large language models reduce public knowledge sharing on online Q&A platforms

#343

Post by zahlman »

verdverm wrote: post #69
rkncland wrote: post #31

Earlier quoted context omitted.

ChatGPT plagiarizes the anwers of those whom you call "arseholes". How is using Stackoverflow in read-only mode different from using ChatGPT? Except of course that reading Stackoverflow directly has better retention rates, better explanations and more in-depth discussions. (My view is that moderators can be annoying but the issue is overblown.)

Plagiarizing means violating copyright, loosely speaking. When you, as a human, use SO, you assign your rights to the content to SO. That company is licensing the content to 3rd parties, including those who want to train their LLMs. What I find is that the LLMs are not spitting out SO text word for word, as one would when plagiarizing. Rather, the LLM uses the context and words of my question when answering, making t…

>When you, as a human, use SO, you assign your rights to the content to SO.

No, you don't. You license it. The community gets a Creative Commons license (https://stackoverflow.com/help/licensing), and the company gets a few additional rights (https://stackoverflow.com/legal/terms-of-service/public#lice...). You retain copyright.

Re: Large language models reduce public knowledge sharing on online Q&A platforms

#344

Post by zahlman »

gwern wrote: post #339

Note for those suffering from deja vu: this was extensively highlighted and touted by critics back in July 2023 as "Are Large Language Models a Threat to Digital Public Goods? Evidence from Activity on Stack Overflow" https://arxiv.org/abs/2307.07367 including HN: https://news.ycombinator.com/item?id=36763718

Worth noting Stack Overflow's own meta discussion of that: https://meta.stackoverflow.com/questions/425651 ; relatedly https://meta.stackoverflow.com/questions/425635 and https://meta.stackoverflow.com/questions/422392.

Re: Large language models reduce public knowledge sharing on online Q&A platforms

#345

Post by famouswaffles »

zahlman wrote: post #331
Terretta wrote: post #279

People don't post where they aren't visiting. They're not visiting Stack Overflow for well-read material (the popular languages) because perplexity.ai, ChatGPT, Claude, etc., not only answer questions better than reading pages of StackOverflow, but will cut and paste the (right or wrong) answer for you faster. If you're not on StackOverflow while asking questions, you aren't answering things there either. Nothing els…

Competing with AI is an explicit non-goal of the Stack Overflow community, as well as the one other Q&A forum I use (Codidact). AI won't "answer questions better". It will cut out the middleman of parsing your question and matching it up with words in the shape of an answer. It will also frequently hallucinate, and essentially never sanity-check what you're trying to do. But the main thing giving it a speed/convenien…

It definitely answers certain questions better and languages, topics these models are less capable at are hit less. The denial at this point doesn't make any sense.

Re: Large language models reduce public knowledge sharing on online Q&A platforms

#347

Post by famouswaffles »

rytis wrote: post #232
elicksaur wrote: post #155

Earlier quoted context omitted.

On the other hand, I find it to be a fatal flaw that LLMs can’t say, “Hey you probably don’t actually want to do it that way.”

I think it depends on how the question is constructed: - I want to do X, how do I do it? - I was thinking of doing X to achieve Y, wonder if that's a good idea? Sometimes, I really want to do X, I know it may be questionable, I know the safest is "probably don't want to do it", and yet, that's not someone else's (or LLMs) business, I know exactly what I want to do, and I'm asking if anyone knows HOW, not IF. So IMO i…

I think there can be a middle ground. I think it's fine if LLMs warn you but still answer the question the way you asked. I don't always know when i should be asking if something is a good idea or not.

Re: Large language models reduce public knowledge sharing on online Q&A platforms

#348

Post by CptFribble »

nfw2 wrote: post #287
ilrwbwrkhv wrote: post #277

Earlier quoted context omitted.

Curious about your complex use case of epub.js. What were you trying to do with it?

I'm building an e-reader app where "enhancement content" such as illustrations, context-approprate summaries, and group chat can be integrated into the reading experience. The way I am connecting external content to the epub is through an anchoring system -- sequences of words can be hashed to form unique ids that are referenced by the enhancement. Doing this lets me index the enhancement content in such a way that i…

> group chat

hmm having a sort of mini-forum-like experience tied to particular pages in a book seems like a fascinating idea! being able to discuss plot twists and such only once you've already gotten to that point?

wow this seems like an amazing idea actually! any names yet? I'd love to check it out once it's done!

Re: Large language models reduce public knowledge sharing on online Q&A platforms

#349

Post by zmgsabst »

roughly wrote: post #143
zmgsabst wrote: post #100

Earlier quoted context omitted.

You can usually even ask the same LLM: - do a task - criticize your job on that task - redo that task based on criticism I find giving the LLM a process greatly improves the results.

What’s fun is that you can skip step 1. The LLM will happily critique its own nonexistent output.

So?

I too can write made up criticism if that’s what my boss wants in the workplace — but that doesn’t suddenly invalidate my ability to criticize my own work to improve it.

Re: Large language models reduce public knowledge sharing on online Q&A platforms

#350

Post by haolez »

oniony wrote: post #317
haolez wrote: post #269

Earlier quoted context omitted.

To be fair, after the ban expired, I started submitting the test cases as instructed and the community was very helpful under these constraints.

Are you saying it worked?

Well, the tone was unnecessarily rude, and the ban period was excessive, but after complying, they were very helpful.
Post reply on HN