160

I'm a new contributor trying to get more into helping on Stack Overflow, and after spending a while writing up an answer, it was deleted because it seems like I used ChatGPT.

It might have been a mistake since after I answered it the asker edited the post to ask a follow-up question to my answer, so it looked like I didn't properly answer the question.

Sorry if it sounds like I'm making a big deal out of nothing, but it's very annoying after I spent my own time helping someone else, just for someone else to delete it all and accuse me of writing it with a robot.

Here's the link - as it is deleted I'll add a screenshot also. https://stackoverflow.com/a/74754627/14769987

screenshot of the deleted answer

25
  • 33
    Did you try asking in a flag rather than meta?
    – Nick is tired
    Commented Dec 11, 2022 at 22:31
  • 46
    There will be bumps in the road with the chatGPT ban -- that's to be expected, but overall, the ban was the right call. Your case is one of those bumps in the road that the moderators should be able to remedy. (and good for your for following through with the issue)
    – David C. Rankin
    Commented Dec 12, 2022 at 5:11
  • 40
    Maybe we can change the requirements for the Not a Robot badge. Sounds like you deserve it... ^_^
    – Tomerikoo
    Commented Dec 12, 2022 at 7:59
  • 38
    This is essentially going to be the real question. Not "How do we recognise ChatGPT answers", but "how do I make my own answers not look like ChatGPT". Man that AI is going to make life miserable.
    – Gimby
    Commented Dec 12, 2022 at 8:58
  • 41
    fortunately, after the first test phase it's going to become a commercial service. No freebies to post trash then.
    – Jean-François Fabre Mod
    Commented Dec 12, 2022 at 9:54
  • 29
    @Jean-FrançoisFabre wow, an actual example of where commercialism saves the day. I hope greed is also involved here, so it happens sooner rather than later :)
    – Gimby
    Commented Dec 12, 2022 at 10:01
  • 6
    @NickstandswithUkraine how to ask in a flag?
    – Konrad
    Commented Dec 12, 2022 at 10:19
  • 10
    All I can say is please error on the side of caution. Make sure something is generated by a bot before marking it as such. If in doubt, assume it was written by a human. This is supposed to be a welcoming community!
    – JosephDoggie
    Commented Dec 12, 2022 at 13:35
  • 14
    @NickstandswithUkraine Please don't use flags to question/dispute moderator actions, that's not what they are for. Creating a meta support post is the most appropriate.
    – Lundin
    Commented Dec 12, 2022 at 15:15
  • 10
    This really looks like a bot answer.
    – Andrew Allen
    Commented Dec 12, 2022 at 15:29
  • 6
    @Trilarion Overly-formal phrasing ("Rather than", "whereas") and general structure (paragraph of explanation -> "Here's an example" single sentence that's longer than necessary because it duplicates context (mentioning the function again) -> code) are what set it off for me.
    – Izkata
    Commented Dec 12, 2022 at 15:57
  • 15
    @Izkata Just had a look at some of my early answers from almost 10 years ago. I wasted time on sentences like: "So what you have to do is something like", "So my answer would be" and "I now found that it can be done as follows". Not very great writing, but also not a dead-sure bot signature.
    – NoDataDumpNoContribution
    Commented Dec 12, 2022 at 16:23
  • 13
    Simple: You didn't pass the Turing test. Sorry.
    – Peter - Reinstate Monica
    Commented Dec 12, 2022 at 18:15
  • 8
    @Lundin Please do use flags to dispute/question moderator actions, doing at least one follows a clear escalation path, there's every possibility a simple mistake was made and a flag is the best way to handle that. If a flag doesn't result in the action you expect, that's when meta is useful, not as a first stop.
    – Nick is tired
    Commented Dec 12, 2022 at 21:37
  • 7
    After reading through the comments and answer below I can understand why your answer was deleted, but I don't understand why it's still deleted. If the reason for deletion was that the mod thought it was a bot answer, now that we know it's not, that action should be immediately reversed.
    – Eaten by a Grue
    Commented Dec 13, 2022 at 14:50

1 Answer 1

Reset to default
122

We had a "ChatGPT" attack this week. Most of answers were rather generic, and contained that exact sentence start:

Here's an example of how you could use ...

Followed by example/generic code that isn't related at all to OP code (and even gets puzzled comments).

So your answer looks very much like a generated ChatGPT answer (even if some other "dead giveaway" parts are missing) which explains that it was deleted.

I've reasons to believe you because

  • I tried hard to generate an approaching answer with ChatGPT and failed.
  • The answer is within your domain of expertise
  • You didn't flood the site with other generic answers since ChatGPT was made available

But the answer seems not tailored to the OP question at all and may very well have been generated at first sight. It was even flagged "not an answer". Not a great answer to the OP question in the current form.

I suggest that you edit it to make it less generic and more related to the original question, and then flag for moderator attention so we can decide if we undelete it.

42
  • 42
    What is the idea of deleting answers under the argument that they were apparently created by an AI? It makes much more sense to explain that the answer was deleted because it is not related to the question asked or because the answer belongs to a user who has a proven track record of spamming, which is more objective and saves moderators from having to get into a ton of complex discussions. If an answer is not quality (whether it was created by an AI or not) it will easily fall to the bottom.
    – Tedpac
    Commented Dec 12, 2022 at 0:06
  • 118
    @Tedpac: ChatGPT crap creates a much higher moderation burden than manually-written crap, while taking extremely low effort to post. We don't have the expert resources to carefully read through all the AI crap and verify, yup, it's crap again. Plus, low-quality or wrong answers get upvoted and accepted all the time. They don't just "easily fall to the bottom".
    – user2357112
    Commented Dec 12, 2022 at 0:22
  • 51
    @Tedpac please read Temporary policy: ChatGPT is banned. You probably won't believe it if you didn't see it by yourself, but at that time, too many users flooded the site with ChatGPT answers as-is and not caring whether the answer is correct or not (because they actually don't have expertise on that tech stack), and even got accepted by the OP (because they usually don't know better). Wrong answers are not a reason for deletion, AI-generated answers are.
    – Andrew T.
    Commented Dec 12, 2022 at 2:13
  • 21
    Wait, so now mods are deleting answers because they think they're ChatGPT without even verifying? and how is this answer considered "generic code"?
    – java-addict301
    Commented Dec 12, 2022 at 3:11
  • 30
    @java-addict301, how would you propose that they verify? Unless the poster admits to it (which I've seen a few times) there's often no way to know for sure.
    – Chris
    Commented Dec 12, 2022 at 3:56
  • 56
    @Tedpac ChatGPT is a text generator, not a fact generator. It takes seconds for ChatGPT to generate a plausible-looking answer and longer for a human with subject matter expertise to identify the answer as factually wrong. There was an obvious wave of this nonsense recently, mostly copy/pasted by new or low-rep accounts. If you spent time in the review queues you would know that human effort to moderate that wave of junk is grossly outmatched by ChatGPT's ability to generate junk.
    – Blastfurnace
    Commented Dec 12, 2022 at 4:07
  • 36
    @java-addict301, please read Temporary policy: ChatGPT is banned and Why posting GPT and ChatGPT generated answers is not currently acceptable. There are some very good reasons for doing this. For now, it's only temporary.
    – Chris
    Commented Dec 12, 2022 at 4:18
  • 15
    @Chris I have read the policy and I agree, however deleting even one valid answer in the quest to stop ChatGPT is like SO cutting off its nose to spite its face so to speak, and will ultimately do far more damage to the community than allowing ChatGPT answers (IMO). See OP's disposition - "Sorry if it sounds like I'm making a big deal out of nothing, but it's very annoying after I spent my own time helping someone else, just for someone else to delete it all and accuse me of writing it with a robot". This is not sustainable.
    – java-addict301
    Commented Dec 12, 2022 at 4:30
  • 35
    @java-addict301: Due to a hopefully-short-term flood of essentially trolls posting bad answers (or children playing with loaded guns), we unfortunately have to take drastic measures to keep up with it. When there is some unfortunate collateral damage (false positive identification of answers as being ChatGPT), well-intentioned humans can communicate and get it sorted out. This is not ideal, but there simply isn't time for experts and/or mods to spend carefully reading all answers that smell GPT-generated in case there's something useful, instead of doing more valuable with their time.
    – Peter Cordes
    Commented Dec 12, 2022 at 4:40
  • 35
    And yes, all the ChatGPT answers I've seen in the tags I follow (assembly, cpu-architecture, etc.) have been bad, off-topic at best, full of wrong facts at worst. Apparently some people have had success getting the bones of an answer from ChatGPT and hand-editing it to fix bugs in the almost-working code it generated, but the problem is people who aren't doing that, instead posting the raw garbage.
    – Peter Cordes
    Commented Dec 12, 2022 at 4:42
  • 36
    linking to a heavily downvoted answer of yours isn't really useful
    – Jean-François Fabre Mod
    Commented Dec 12, 2022 at 7:45
  • 34
    @Tedpac tell me, did you read 192 answers? In one day, I mean. It's very important to answer because on Sunday, 11th of December 2022 there were 4998 answers. Of course that's just the ones that are still visible - sans all the deleted ones for any reason. But it's still a number we can work with. We have 26 moderators. And I'm being generous, not all are active. But still. 4998 / 26 = 192.23 answers per mod to read. Did you read that many? In one day.
    – VLAZ
    Commented Dec 12, 2022 at 11:19
  • 15
    @Tedpac "regardless of whether a response was written by a human or an AI, a moderator will read it" You are sorely mistaken if you think moderators read all answers. Even non-diamond moderators combined with diamond moderators are only able to view a fraction of the total number of posts on the site. You would need at least an order of magnitude more moderators than there are today, maybe two orders.
    – TylerH
    Commented Dec 12, 2022 at 14:15
  • 16
    @TylerH It seems to me that most of the people whining about this just don't have a clue about how much effort curation of this site takes (let alone how much is actually needed). A quick look at the voting activity in their profile usually says it all. Just in the case of the user you're responding to... 50-odd votes in 4 years, 3 of those downvotes...
    – Dan Mašek
    Commented Dec 13, 2022 at 1:00
  • 24
    the debate about chatgpt is done. That ship has sailed. No need to debate the usefulness of chatgpt answers here.
    – Jean-François Fabre Mod
    Commented Dec 14, 2022 at 9:13

You must log in to answer this question.

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