25

I’m NOT suggesting a network wide ban - that already has a separate post.

I recently noticed that a user posted a ChatGPT generated answer that totally failed to answer the question that was asked (and they didn’t mention the use of ChatGPT). I custom flagged it, but I looked back at MSE, and since I don’t think it’s been discussed yet:

Should we ban ChatGPT answers on MSE?

I tend to lean towards banning it as there is a rather high chance of it being wrong or failing to answer the question

4
  • 9
    This one's actually pretty funny, in comparison to most ChatGPT bans network-wide, which are on a factual / ethical basis. If someone uses ChatGPT to make a [discussion] point about how the network should operate, and it expresses an opinion other people agree with & upvote, should it be ignored? On the one hand, obviously take it seriously, but on the other hand, obviously ignore it... Just playing around, not seriously advocating it should be allowed. (I don't think it would be a good idea to allow it, but alas, I'm somehow still not the appointed dictator of Meta Stack Exchange.)
    – Slate StaffMod
    Jan 11, 2023 at 6:51
  • 5
    @Slate I wonder how many people would agree with a point when it is clearly attributed to ChatGPT, and not plagiarised by a user just copy pasting it without any attribution.
    – Luuklag
    Jan 11, 2023 at 7:35
  • In my opinion, instead of banning such answers, it would be better to REQUIRE that the author cites ChatGPT as the source of the answer. Jan 12, 2023 at 12:34
  • 1
    @JohnMiliter this solves basically nothing of what the major concerns are. That the generated content is completely unreliable, yet presented in a way to convince you they are correct. The goal of the network is to provide reliable content. People can already go get an answer by ChatGPT, if that's their fancy. Reposting the content here serves no useful purpose - it can even be viewed as an attempt to borrow legitimacy by being present on the network. And in turn it can hurt the network's legitimacy because the answers are not actually correct.
    – VLAZ
    Jan 12, 2023 at 14:47

2 Answers 2

27

I say ban them (which they already are treated like, we've deleted quite a few already as moderators). As you mentioned, on the objective questions here those answers can be wrong (often are), so the same policy as on Stack Overflow applies.

On the subjective questions, well... Ideally, Good Subjective answers are written from experience. Something is likely informing your opinion. Don't just use the AI to regurgitate an opinion for you, write down your actual opinion and why it is the way it is. In that sense, this quote from the SO policy still applies:

GPT risks breaking readers’ trust that our site provides answers written by subject-matter experts.

The subject-matter experts in this case being users with an informed opinion on anything discussed here, because they have their own experiences with the problem that's being discussed. Those are the kinds of opinions I expect to encounter on posts here, not the ones generated by our evil AI overlords.

15

Well, practically speaking it is banned to some extent. Some individual thinks they are clever and posts a reply that clearly is generated by ChatGPT and gets suspended for posting nonsense. There's no 'formal' rule because don't post crap is obvious.

I think there's an important point to be made that meta supports the network as a whole, and as such, well posts need to show understanding of the question, context and sometimes history of the network.

So far, a good chunk of ChatGPT posts we've run into fall into 3 broad categories - they tend to be new users with no apparent activity elsewhere, folks answering the question on how to detect ChatGPT or spammers. It’s also invariably 'new' users or at least throwaway accounts.

And frankly I've not seen any actual 'considerate' use of the tool, nor any real awareness of the network in these posts.

Fundamentally, the way it shows a certain lack of effort, respect for the commons, is essentially plagiarism and frankly is folks trying their luck.

When we do suspensions - here's a strange thing, the users rarely respond. I got an apology on Pets where I warned someone. Here? I can suspend someone a year and... no reply. It’s almost like they don't care, or want to find out if they can get away with it.

With nearly all these being 'new' users, I also feel like there's no good intent being demonstrated here.

Let’s say, hypothetically - someone trained a ChatGPT type tool on - well meta. Maybe the collected works of Shog9. It answers questions perfectly with awareness of the actual question and doesn't give generic prepackaged dog chow. I'd certainly be impressed - and consider it useful. So far that has not been the case.

On the other hand, the poor quality, lack of effort and the noise these posts create mean they don't add value and should be discouraged.

Basically, I'd say 'don't try your luck' if you want to do it, and we reserve the right to deal with folks trying their luck as we deem appropriate.

1
  • "need to show understanding of the question, context and sometimes history of the network." That sadly is the most useful one. Why certain things are the way they are, going to the first entry, rather than inventing a reason post-facto. There are multiple "rules" that people explain away with stuff literally took out of their asses, yet when you try to educate them why it's like that, they rebuff you because revisionism.
    – Braiam
    Jan 12, 2023 at 14:28

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .