38

Edit: This question is old. Please see the newer post:

Use of ChatGPT is now banned on Super User


There is a new AI chatbot called ChatGPT by OpenAI. It appears to be able to answer questions and generate code based on what you ask it.

It has already been banned on Stack Overflow. They even have a banner on their site pointing to the new policy.

ChatGPT notice banner

There is also a post on Meta asking for it to be banned sitewide.

It has now arrived on Super User: https://superuser.com/a/1757485/403203

How do we deal with this and answers generated by it?

5
  • 7
    At an absolute minimum any ChatGPT answer should be treated as if you found it at an external source. Including disclaimers that it is not your work and 100% blockquoting the section of copied content following the guidelines at How to reference material written by others. It is not your work, you are essentially plagiarising a bot. That said I am definitely siding with an outright ban of these kinds of answers.
    – Mokubai Mod
    Dec 13, 2022 at 8:10
  • 13
    If I want answers from ChatGPT, I can just ask ChatGPT. So yeah, it has no place here IMHO.
    – Daniel B
    Dec 13, 2022 at 12:33
  • 5
    @DanielB but if you want correct answers from ChatGPT, you need ChatGPT + a human to check the answer. Dec 14, 2022 at 15:19
  • 2
    @Giacomo1968 Good question. I assume moderators have tools that can help them. Apparently they can tell
    – MC10
    Dec 15, 2022 at 5:23
  • 1
    As of 24/12/2022 the highest voted answer by a big margin is that by forest which is in favour of a ban. As a result going forwards AI generated questions and answers are not allowed on Super User and will be removed with no prior warning. In addition, users may be suspended or deleted.
    – DavidPostill Mod
    Dec 24, 2022 at 7:23

6 Answers 6

56

Yes

While it may be fine in theory to allow any answer as long as it is correct, what we're seeing across the network is that the answers may appear good on the surface but contain subtle (or not-so-subtle) inaccuracies. It should be banned here for the same reasons that it is on Stack Overflow.

7
  • 3
    Fully agree, forest. I guess my sample answer can be considered a "limit test" of sorts. I can only imagine the complications it would cause with moderation with widespread use. Although... how to detect ChatGPT answers (that aren't a strict copy paste job) may be another problem.
    – MC10
    Dec 13, 2022 at 22:35
  • 6
    I had to flag around a dozen answers from a single user who was submitting ChartGPT content as an answer. The general theme is every single answer had an ordered list of steps. The problem with a handful of those answers, is they were actually solving the opposite problem, one in particular wanted something always enabled and the answer provided steps to disable that feature. User when presented with the comment they were using ChartGPT to generate content just deleted their answers. It appears a moderator nuked that user from orbit since 98% of their content was low quality ChartGPT content
    – Ramhound
    Dec 14, 2022 at 18:46
  • Ramhound, how did you recognize ChartGPT content? I hear about ChartGPT for the first time...
    – r2d3
    Dec 17, 2022 at 23:57
  • 1
    @r2d3 I think Ramhound mostly answered that. ChatGPT content is recognizable because (a) the style is typically the same "ordered list of steps" that ChatGPT favors, (b) often wrong, (c) posted in rapid succession. There can be other indicators that might make me suspicious, such as answers on a wide variety of different topics (most users tend to answer within a smaller range of topics), etc. Dec 19, 2022 at 23:17
  • @Ramhound Many of my answers are ordered or bullets lists. Does it mean that my "style" is ChatGPT-ed or did ChatGPT followed mine :) ? Dec 25, 2022 at 14:48
  • @ReddyLutonadio - I have never seen one of your answers. I have linked to two of the deleted user's answers though, compare and contrast.
    – Ramhound
    Dec 25, 2022 at 21:57
25

Yes.

ChatGPT is only Google on steroids, parroting stuff with no understanding or experience. It will copy bugs, and its generated code, while it may work for the test values, may fail for others, or may contain known security bugs.

We really expect answers to be based on experience, or to be found in some credible source, or that have been critically examined by the person to post the answer. For ChatGPT, we have none of these.

Answers by ChatGPT are at best a collection of random fragments and cannot really be called "answers".

1
11

Yes.

But:

  • ChatGPT or similar technology might be really good for helping to find duplicate questions, or a probability score that the question has been asked before.

  • It would be interesting if ChatGPT or other AI itself (not via human copy-paste) was basically an occasional random contributor on a leash.

    • Perhaps: such contributions initially only visible only to users with enough rep and placed in review queues, and if judged to submit good answers, then they'd show up. Question asker would still have to accept it.

    • If a long time passes and AI sucks, that's good data to have, right?

1
  • Let's start by having the question search act properly, then think about ChatGPT integration...
    – einpoklum
    Jan 1, 2023 at 21:58
3

Certain uses of ChatGPT should not be allowed, but a "total ban" of all uses is a bit overboard.

Straw-man proposal (reposted from my meta.askubuntu.com answer on the same topic - More info there; short-version here):

  • ChatGPT should not be used to write answers, but rather to improve answers.

  • A ChatGPT-based answer should never be posted by someone without subject matter expertise in the question being asked in the first place. Confirming that the answer works is not enough. If you are even using ChatGPT to assist you in crafting the answer, then you must have the ability to:

    • Confirm that the answer is correct and works
    • Fully understand the answer that you are posting
    • Correct any issues
    • Communicate the answer in your own words
    • Reply to follow-up questions/comments about your answer
    • Warn of corner cases or potential issues if needed
    • Cite any reliance you made on ChatGPT for the answer

That's in addition to the normal requirements for a good answer. As just one example from that help page, answering a commonly-asked-duplicate is not useful, regardless of whether your answer is AI-assisted or not.

17
  • 1
    “Certain uses of ChatGPT should not be allowed, but a "total ban" of all uses is a bit overboard.” - A total ban is the only way to prevent improper usage of the tool. I haven’t seen a single Chart GPT answer that actually answered a SU question and I have seen (and reported) more than a dozen.
    – Ramhound
    Dec 20, 2022 at 1:19
  • 1
    @Ramhound Just to be clear, in advocating for a "total ban", you seem to be saying that you or I (both experienced users) should never use anything found via ChatGPT in an answer or edit here on Super User, right? Dec 20, 2022 at 1:24
  • 4
  • 1
    @Ramhound And in my experience (mentioned in the linked Ask Ubuntu answer) it has generated information that has improved answers and posts here on Stack Exchange (and elsewhere). I'm not a fan of banning a useful tool. Instead, I propose that we allow users who can to use it responsibly, and ban users who don't use it responsibly. Dec 20, 2022 at 1:34
  • 1
    @Ramhound I'm also confused on why you would cite that answer as a reason why a "total ban" is necessary. Under the policy I proposed, the use that you cited in that linked would have been disallowed anyway. A total ban is not necessary to prevent that. Dec 20, 2022 at 1:39
  • A policy that allows some users to use Chart GPT and disallows others from using it will be impossible to enforce. Besides every single answer I have seen Chart GPT used to answer a question has been beyond horrible. I don’t see any use out of such a horrible tool. I personally won’t have my opinion swayed by the usefulness of any answer that contains CGPT content. Did you read either of those deleted answers?
    – Ramhound
    Dec 20, 2022 at 6:48
  • 1
    @Ramhound I did read them. Did you read my answer above? Both of the answers you cite would have been banned under the first bullet point on my proposal, without requiring a total ban -- "ChatGPT should not be used to write answers". This is the only usage that is going to be detectable anyway, so why bother trying to ban other good uses which you aren't even going to be able to detect? I've shown in three examples now that there are good uses of the tool when used responsibly. We're in full agreement that bad usage should be banned, but that should not preclude good usage. Dec 20, 2022 at 7:11
  • 1
    @Ramhound "I personally won’t have my opinion swayed by the usefulness of any answer that contains CGPT content." Then I'm sorry, but how can you expect anyone here to take your position seriously if you state outright that you won't be swayed by demonstration of its usefulness? Dec 20, 2022 at 7:16
  • Your grammatical corrections, were trivial, nothing unique about those improvements were specifically unique to CGPT. Been using grammatical addons for years.
    – Ramhound
    Dec 20, 2022 at 7:25
  • 2
    @Ramhound (a) I disagree that they were "trivial" -- One of those was almost unreadable without extreme effort (you would have VTC'd based on your previous comment), and (b) I challenge you to find any non-AI "grammar tool" that could make that improvement. Dec 20, 2022 at 7:35
  • 2
    But assuming you are right, why ban the use of ChatGPT for this purpose? Why ban ChatGPT and yet not other "grammar addons"? And how would you detect that it was being used for that purpose in the first place if it wasn't disclosed? Dec 20, 2022 at 7:35
  • I feel like this banning of ChatGPT is the first step towards Luddism. If the answer is a working answer it can be voted up. If it is a bad answer it can be voted down. There may be fools that try to work ChatGPT for points and ChatGPT can indeed produce some bad answers (at THIS time), so I understand the skepticism. But a full and permanent ban is not only going to be impossible, it is throwing the baby out with the bath water. Feb 6, 2023 at 17:20
  • @BrianMcMahon - You can’t see the answers that have been flagged then deleted. The reason these AI written answers are not allowed, or shouldn’t be allowed, is due their quality. I have flagging hundreds of these answers, not a single one was properly cited, nor did they even answer the question that was asked.
    – Ramhound
    Jun 5, 2023 at 13:06
  • @Ramhound Not a single one? I guess I've flagged more (over 1,200 so far across three sites), but I've definitely seen those that answered the question correctly. They were in the minority, which is certainly problematic. I've also seen those that cited ChatGPT (or Bing Chat) usage as well, but under the new "Don't ask; don't tell" policy, those would be eligible for deletion while those that plagiarized (didn't cite) would be difficult to remove. So right now, users are incentivized to not cite. Jun 5, 2023 at 15:18
  • @NotTheDr01ds - The answers I have seen answer questions that were not even part of the question that was asked. This is explained by the fact the learning language model can't understand the context. I personally will continue to find contributions generated by OpenAI or similar language learning models as unhelpful.
    – Ramhound
    Jun 5, 2023 at 15:38
0

I'm in favour of banning ChatGPT from Superuser (and many other StackExchange) forum, but please refrain from being too harsh:
Recently, I was doing reviews, I agreed on a "First Answer", but apparently that answer was auto-generated by ChatGPT, so my evaluation was wrong.

No problem there, but this question seems to be used as a review test, and that goes too far: review tests can contain rubbish, spam, insults, ..., but a Superuser post, containing correct information, using as a review test because the author is (believed to be) a computer and not a human being (which will become more and more difficult), that's a bridge too far and will only reduce the amount of reviewers on this site.

Hence, I do agree deleting ChatGPT post from this site.
I do not agree using such posts a review tests.

In case somebody says "This is not an answer on the question whether or not to ban ChatGPT from the site", my reaction is "It is my way to agree to the statement and to add information on how to proceed while doing that".

1
  • 2
    The answer you reviewed didn't contain accurate or helpful information and was clearly generated by an AI, so it's actually a perfect candidate, to verify reviewers are able to properly identify helpful answers. The answer contained generic information, didn't directly answer the question, that is one way you can tell it was generated by an AI.
    – Ramhound
    Jun 9, 2023 at 15:16
-10

There are several compelling reasons why learning to use AI (Artificial Intelligence) can be beneficial:

  • Problem-solving capabilities: AI provides powerful tools and algorithms that can analyze vast amounts of data, identify patterns, and make predictions. By learning to use AI, you can leverage its problem-solving capabilities to address complex issues and make more informed decisions.

  • Automation and efficiency: AI can automate repetitive tasks and streamline processes, saving time and resources. By learning to use AI technologies, you can optimize workflows, increase productivity, and focus on more strategic or creative tasks.

  • Innovation and competitiveness: AI is transforming various industries, and organizations that embrace AI have a competitive advantage. By acquiring AI skills, you can contribute to innovation within your field, develop new products or services, and stay ahead of the curve.

  • Data-driven insights: AI excels at extracting valuable insights from large datasets. By understanding AI techniques, you can harness the power of data to gain meaningful insights, discover trends, and uncover hidden patterns that can drive business growth or inform decision-making.

  • Enhanced decision-making: AI algorithms can analyze complex information and provide recommendations, aiding decision-making processes. Learning to use AI empowers you to leverage these algorithms to make more accurate and data-driven decisions, reducing errors and improving outcomes.

  • Personal and professional growth: AI is a rapidly evolving field, and acquiring AI skills can open up new opportunities for personal and professional growth. It equips you with valuable expertise that is in high demand across various industries, making you a sought-after professional.

  • Addressing societal challenges: AI has the potential to tackle pressing societal challenges, such as healthcare, climate change, transportation, and more. By learning to use AI, you can contribute to finding innovative solutions and make a positive impact on society.

Overall, learning to use AI offers a wide range of benefits, from increased efficiency and competitiveness to fostering innovation and addressing societal issues. It empowers individuals and organizations to leverage data-driven insights and make informed decisions in an increasingly data-rich world.

12
  • 3
    If you're going to post an AI-generated response, then at least give attribution, because the current CoC... explicitly disallows copy-pasting without attribution: "Plagiarizing or copying content from websites, books, or other online and offline tools without proper attribution in a manner that violates our referencing standards."
    – Andrew T.
    Jun 4, 2023 at 5:01
  • @Andrew T.: I see no mention of ChatGPT or AI in the links you posted. The use of AI is not universally considered plagiarism. "The definition of plagiarism is the act of using someone else's work or ideas without giving proper credit to the original" ChatGPT or any other AI is not someone else and it is not a person, it is a tool. Jun 4, 2023 at 6:02
  • 2
    As you mentioned, it is a tool: "other online and offline tools without proper attribution", and a staff has also mentioned it explicitly about attribution for LLM, where it seems you have at least noticed that post before.
    – Andrew T.
    Jun 4, 2023 at 6:07
  • Does "other online and offline tools" include computers and typewriters? Who makes this stuff up? under European (and US) law AI cannot own copyright, as it cannot be recognised as an author and does not have the legal personality which is a pre-requisite for owning (intangible) assets. Thus it is not subject to plagiarism. Jun 4, 2023 at 6:19
  • "For now, it seems that artists / creators who use AI to support their creative process may be able to claim ownership of the work if it reflects their choices and creativity". - European Commission Jun 4, 2023 at 6:26
  • Have you been following the subject lately? SE has changed the policy about AI recently: meta.stackexchange.com/a/389583/495084 and all the moderators have gone on strike: openletter.mousetail.nl Jun 4, 2023 at 6:38
  • 3
    Yes, they ban the usage of AI detector and heuristic/guess-based moderation, but still requires attribution for AI-generated content.
    – Andrew T.
    Jun 4, 2023 at 6:45
  • 2
    @C.S.Cameron - I have indeed. I will continue to downvote contributions that clearly contain content generated by OpenAI or similar language learning models. Furthermore, plagiarist have no place in this community, plagiarism is typically a habit.
    – Ramhound
    Jun 5, 2023 at 2:55
  • @Ramhound: AI is just a tool, a computer program. The European Commission has said: "For now, it seems that artists / creators who use AI to support their creative process may be able to claim ownership of the work if it reflects their choices and creativity". Jun 5, 2023 at 8:00
  • Cornell Law School states: "Plagiarism is the act of taking a person's original work and presenting it as if it was one's own" Do you consider AI as a person with human rights? The Supreme Court says not. It must be the user of a tool that owns what the tool creates. (Unless the work is created for pay). Jun 5, 2023 at 8:14
  • 1
    You are oddly fixated on what the Supreme Court and European Commission have to say, while ignoring what has been said on this network as it pertains to your answer: you still have to indicate when you have used a tool (online or offline) to create your post
    – bertieb
    Jun 5, 2023 at 10:07
  • 3
    I don't see how this 'answer' even begins to address the question raised in OP. It mainly argues AI's strong points in general and it reads like an advert for AI/ChatGPT etc. but it's not an answer to question asked. If this answer was generated to by AI, it IMO illustrates why it has no added value for this website. Jun 5, 2023 at 10:10

You must log in to answer this question.

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