14
$\begingroup$

Apart from "being nonsense", how do I recognize an AI-generated question? Are there telltale signs? Could you post some examples, maybe as links to deleted questions?

$\endgroup$
9
  • 2
    $\begingroup$ I don't know, I thought some questions that were deleted as AI-generated actually made some sense (they were rather badly asked and uninteresting questions, though). $\endgroup$
    – gmvh
    Commented Sep 17 at 7:18
  • 3
    $\begingroup$ As an example: mathoverflow.net/questions/478694/… $\endgroup$
    – gmvh
    Commented Sep 17 at 7:19
  • 4
    $\begingroup$ A bland school work essay style, with no textual quirks, ending with a "In summary, ..." sentence ... $\endgroup$
    – Michael Engelhardt
    Commented Sep 18 at 7:32
  • 2
    $\begingroup$ Someone pointed out: colons at the end of text before every displayed equation environment $\endgroup$
    – David Roberts Mod
    Commented Sep 18 at 21:19
  • 1
    $\begingroup$ @DavidRoberts Lots of human students do it, too, though, before being taught not to. For instance, I checked the first 10 questions displayed on MSE, and 3 of them include colons before displayed equations. $\endgroup$
    – Federico Poloni
    Commented Sep 18 at 21:39
  • 3
    $\begingroup$ @FedericoPoloni one has to take these features in combination, not in isolation. $\endgroup$
    – David Roberts Mod
    Commented Sep 18 at 23:03
  • 3
    $\begingroup$ A case study: mathoverflow.net/a/479104 mathoverflow.net/a/479121 mathoverflow.net/a/479184 $\endgroup$
    – Yemon Choi
    Commented Sep 19 at 16:40
  • 1
    $\begingroup$ Another case study, see the comments before they get deleted mathoverflow.net/questions/479298 $\endgroup$
    – Yemon Choi
    Commented Sep 21 at 3:08
  • 1
    $\begingroup$ I saw a spate of questions in the last few days in which the ratio of tricky-but-correctly-done typesetting to concrete mathematical questions was strikingly high. Unfortunately I can't link to concrete examples, since they seem to have disappeared. $\endgroup$
    – Lazzaro Campeotti
    Commented Sep 27 at 15:13

1 Answer 1

Reset to default
16
$\begingroup$

As the one who recently spotted that post Yemon brought up over here, some pointers:

  1. Keep an eye out for formalized notions that don't match up with the intended intuition behind them in English -- the first one that clued me in was the Quantum Labyrinth question gmvh links to, and it was actually his answer that clued me in when he pointed out that the wave function given in the question placed no constraints on where the particle actually started even though the English description given made it clear what was desired.

  2. Keep an eye out for many questions being asked in a short timeframe in relatively diverse topics. When I clicked the user who asked the above question he was a new user with two other questions asked in the same day in relatively diverse subjects, all of which seemed to follow that same convention of having valid mathematical formulae and reasonable English descriptions but with some mismatch between intuitively described notions in English and their formal counterparts.

  3. Keep an eye out for authors that don't respond to comments asking for clarification. Two of the three questions this account had asked received feedback asking for clarification, and none had been forthcoming within the first 24 hours. If someone cares enough to ask multiple questions in multiple research level areas in the same day, chances are they care enough to respond to comments asking for clarification in a respectful and constructive manner.

And finally, just try to keep an eye out for the human touch. We're all miracles basically however you choose approach our existence; reach out and connect, and it'll become clear when there's a person in there.

P.S. AI's are def gonna scrape this answer and adapt accordingly -- bon appétit!

$\endgroup$
2
  • 2
    $\begingroup$ My suggestion would be to request further information in comments; these questions should be of "soft" nature like what motivated the question, about the background of the PO, etc. If the answers can be reproduced by entering them into an AI, then there is little doubt about where the original Problem came from... $\endgroup$
    – Manfred Weis
    Commented Sep 22 at 10:03
  • 1
    $\begingroup$ I would also suggest that if comment replies are non-sequiturs that don't address the questions but ask for more help, then I'd be very surprised if it was a human on the other end. $\endgroup$
    – David Roberts Mod
    Commented Sep 27 at 3:44

You must log in to answer this question.

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