15
$\begingroup$

Another meta question raised the issue of AI-generated (by ChatGPT or similar) posts.

I have a more specific question. If we see a post that we suspect to be of this genre, would the moderators like us to flag it?

$\endgroup$
8
  • 10
    $\begingroup$ Yes, please. -- If you see such post, then please flag it and mention your suspicion. $\endgroup$
    – Stefan Kohl Mod
    Dec 24, 2022 at 17:08
  • 1
    $\begingroup$ Why flag ChatGPT posts (a type of plagiarism) but not other posts plagiarized (by memory) from standard textbooks? Yes, I would like to a ChatGPT post so identified by the one posting it. And similarly I would like proofs paraphrased from textbooks to have their source identified, even if it is just "I saw this back in graduate school". $\endgroup$ Dec 24, 2022 at 21:24
  • 15
    $\begingroup$ @GeraldEdgar I thought the main problem with the ChatGPT posts was not so much plagiarism as that they are largely gibberish. OTOH, I'm not sure I understand what you mean by "plagiarized (by memory)". It sounds really broad and easily misapplied. $\endgroup$
    – Todd Trimble Mod
    Dec 25, 2022 at 3:19
  • 2
    $\begingroup$ @ToddTrimble Of course gibberish should be flagged however it was written. $\endgroup$ Dec 25, 2022 at 6:47
  • 4
    $\begingroup$ I too don't see here any qualitative difference between ChatGPT gibberish and any other kind of gibberish. From my experience, it is out of the question that ChatGPT could be capable of disguising it in any way here. Maybe not in mathematics, maybe in near future, but not here and now. Until then, I don't think any special measures will be needed. Yes, I am aware that two of its answers have been accepted. I still do not see any qualitative difference from "ordinary" such situations, which also can happen. $\endgroup$ Dec 26, 2022 at 7:28
  • 1
    $\begingroup$ Gerald Edgar's comment about plagiarism seems to envisage a situation in which ChatGPT's answer is so good that it's unethical to pass yourself off as being as mathematically competent as ChatGPT, when you aren't. That could be an issue in poetry or art competitions but it's not an issue on MathOverflow (yet!). $\endgroup$ Dec 29, 2022 at 15:14
  • 2
    $\begingroup$ Has there ever been a post (answer or question) that is correct and reasonably concise but still has a ChatGPT feeling to it...? $\endgroup$ Dec 30, 2022 at 8:21
  • $\begingroup$ @TimothyChow The upside would be that if one were to get sued, just consult the (surely upcoming) LawyerGPT $\endgroup$
    – Hank Igoe
    Jan 3 at 15:39

1 Answer 1

14
$\begingroup$

The question was answered in the comments, but just to underscore: the answer is yes. If you see a post you suspect is AI-written nonsense, flag it and describe your suspicion in the flag comment.

$\endgroup$
4
  • 4
    $\begingroup$ A small clarification: Should we flag them as “Very low quality” or “Needs moderator attention”, or are both OK? (Personally I feel more comfortable using “very low quality”, since usually the fact it’s AI-based is just conjectural, whereas the fact it’s very low-quality is a judgement I’m much more confident in making.) $\endgroup$ Dec 28, 2022 at 17:34
  • 1
    $\begingroup$ @PeterLeFanuLumsdaine Both are okay. $\endgroup$
    – Tim Campion Mod
    Dec 28, 2022 at 19:18
  • 1
    $\begingroup$ Choosing "needs moderator attention" does not commit you to a declaration that the post is AI-generated. $\endgroup$ Dec 29, 2022 at 15:06
  • 1
    $\begingroup$ Does the flag remain even if the poster deletes their post? (I'm thinking of mathoverflow.net/a/437506/8133. Evidence for AI: language patterns, bizarre elementary mistake, citation to texts that don't exist, and the user has been banned at MSE for the same thing(!).) $\endgroup$ Dec 30, 2022 at 6:31

You must log in to answer this question.

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