-9
$\begingroup$

Yesterday (June 6), somebody posted this answer (now deleted, so users with <10k reputation will not see it). I have not bothered checking whether it is correct, but it surely looks reasonable, and is surely not AI-produced. I cannot see the up- and down-votes, but I can see that the total was 0. There are no comments under it. So probably a "meh" answer, but definitely not a fraudulent one. Therefore, now comes the weird part: the answer got deleted by the "Community" user, and the user got destroyed!

I find this very brutal. Could the moderators verify the events and offer some explanations, please? I am afraid only that the heuristic used by "Community" (which is an entirely algorithm-driven user) might have made an innocent victim.

$\endgroup$
9
  • 6
    $\begingroup$ The 'Community' user didn't destroy the account or delete the user. It was a human decision, and while I wasn't the person responsible, you should know that this user had other posts. The answer you link to has very odd non-sequiturs and repetitious wording, as well as sudden flip-flops in notation, and also some argumentation that I think is outright wrong, though the wording superficially reads ok. Also, the "proof by contradiction" that actually just proves the thing outright is a very odd thing to write, it's the kind of thing you see in undergrad-level proofs. $\endgroup$
    – David Roberts Mod
    Commented Jun 6 at 10:41
  • $\begingroup$ While the proof might be (at least partially) wrong, why would "Community" delete it? Let me clarify my question: what was the part that "Community" played in this? I want to make sure that this robot is well harnessed by intelligent human supervisers. In general, users are not to be deleted for writing "meh" answers - downvotes are supposed to deal with this. $\endgroup$
    – Alex M.
    Commented Jun 6 at 11:22
  • 6
    $\begingroup$ I don't claim to understand the inside processes of how the user 'Community' works. It is a placeholder user to which to attach certain actions to, like bumping unanswered questions, anonymous suggested edits, or cleaning up after certain actions by users with superpowers (eg the Stack Exchange staff known as CMs, or MathOverflow moderators). Users with such superpowers are aware of the effects of their actions when it is something like what happened here. I can't say much about this, since it's verging on mod-only territory. $\endgroup$
    – David Roberts Mod
    Commented Jun 6 at 12:40
  • 6
    $\begingroup$ All kinds of actions that are publically anonymized for various reasons are formally attributed to the Community user. See meta.stackexchange.com/a/19739 . $\endgroup$ Commented Jun 7 at 8:55
  • 2
    $\begingroup$ Why do you say the answer was "surely not AI-produced"? $\endgroup$ Commented Jun 9 at 12:29
  • $\begingroup$ @SamHopkins: First, the chaining of the mathematical reasoning seems coherent. Second, "the metric of the images is not greater than the metric of the preimages" - this sloppiness is typical of how a student would speak. An AI system would have probably said "the distance on the image space" (or "on the range"). An AI system would not have denoted a compact metric space by $\mathcal G$ (but rather by $X$, or $M$). AI-generated answers are often elegant and "too perfect"; the slight sloppiness of this one makes it attributable to a human author (undergraduate, I believe). $\endgroup$
    – Alex M.
    Commented Jun 9 at 14:24
  • $\begingroup$ @SamHopkins: Also, the choice to use short sentences suggests someone for whom English is not the mother tongue, and who feels insecure about writing long convoluted complex sentences (like this one). Not how an AI would behave. Also, using $t$ to denote a natural number again betrays a beginner. Believe me, I am pretty good at telling AI from human - everything is in the small details. $\endgroup$
    – Alex M.
    Commented Jun 9 at 14:27
  • $\begingroup$ Your points seems inconclusive to me. I am pretty skeptical anyone can be good at telling AI from human, given how rapidly AI tools are developing and changing. $\endgroup$ Commented Jun 9 at 16:07
  • $\begingroup$ I have got 10/10 in the test found in this article (scroll down a bit, it is an embedded Google form): arstechnica.com/ai/2024/05/…. You have heard it right: 10/10, I am that good! A hint if you want to win the quiz: the AI-generated answers are the ones that feel "too flawless": flawless punctuation, flawless English, flawless moral thinking. And no ego. $\endgroup$
    – Alex M.
    Commented Jun 9 at 16:33

1 Answer 1

22
$\begingroup$

When an account is destroyed (e.g., by a moderator), the posts are labeled as "deleted by the Community user". The label is not a claim that some computer program made the decision to delete the answer.

A number of MathOverflow members seem to disagree with your claim that the answers are "surely not AI-produced", since we got a bunch of flags. Also, the answers are math-shaped nonsense.

$\endgroup$
1
  • 16
    $\begingroup$ Aggressive destroying of nonsense posters is something I am strongly in favor of on science-oriented sites. $\endgroup$
    – Buzz
    Commented Jun 7 at 1:11

You must log in to answer this question.

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