78

As of February 2 2023, there is a ban on the use of AI-generated content on Ask Ubuntu.

This ban involves:

  1. Deletion of all AI-generated content, whether acknowledged as such or not
  2. Moderator warning email for initial occurrence
  3. The usual escalation of penalties for further infractions

Grammar and Spell Checking Tools: Note that it is the moderator team's opinion that the use of Grammarly and spell checking tools to correct grammar errors and spelling errors is not AI-generated content. Using those tools to improve your own written posts' grammar or to correct spelling errors is not forbidden; however, we are specifically barring the use of ChatGPT and other AIs as the sole source of generating answer content.

Timeline:

  • Community Consultation: On December 19th 2022, Community discussion commenced on AU Meta concerning our position on AI-generated content. This discussion attracted considerable Community interest and involvement.
  • Temporary Ban: On January 5th 2023, while Community discussion was taking place, there was a considerable upsurge in the number of of AI-generated answers and spam messages posted on Ask Ubuntu. To get control of this upsurge of incorrect, nonworking, and unvetted content in answers, a temporary ban on AI-generated answers was instituted. This ban continued until February 2, 2023, when a permanent ban was instituted.
  • SE Enforcement Guidelines: On May 29 2023 SE issued a set of enforcement guidelines which prohibited most of the Ask Ubuntu interventions related to AI-generated content. This action, along with several other factors, lead to the general strike by Moderators and curators of SE.
  • Strike Resolution and new Policy: On August 6th 2023, with the strike resolved, a new AI Policy was jointly created by SE staff and Moderators.
18
  • 3
    Is this ban limited to ChatGPT generated answers? Or is using ChatGPT to assist in editing posts also banned? I realize it says a complete ban, but I’m also aware there was some discussion about using it to improve questions (and answers?), hence my question
    – cocomac
    Commented Jan 5, 2023 at 7:15
  • 8
    Ideally discussion will take place on the linked Meta thread where I would be keen to see your input. But: i am not convinced about the use of AI to improve posts in this way. The actual Ask Ubuntu experience with ChatGPT has been very, very different, it has been multiple copy and paste posts that have been for the most part really poor answers. It has been a cool and calculated effort by users to gain rep fast with no interest in accuracy or the production of quality answers.
    – andrew.46 Mod
    Commented Jan 5, 2023 at 7:45
  • 2
    @andrew.46 In light of this announcement, does this thread need any further update/clarification/commentary?
    – Levente
    Commented Jan 5, 2023 at 11:57
  • 2
    @Levente AI grammar checkers are 'limited' in their scope to grammar improvement amd spell checking. This ban is on AI generated content (which is what ChatGPT and others do) not the use of Grammarly AI which in the free grammarly is specific to grammar improvements and spellcheckers.
    – Thomas Ward Mod
    Commented Jan 5, 2023 at 15:52
  • O.k. which flag can I use (that's available to me) to contribute to enforcing this ban? I look at the list of the available flags and it's not totally clear. I did not want to ring "moderator attention" for every single such instance. Do I have another choice?
    – Levente
    Commented Jan 19, 2023 at 17:27
  • 1
    @Levente I'd use a custom mod flag. I typically include why I think it is ChatGPT, the GPT detector result, and any other related info. If it helps, here's a (lightly modified) ChatGPT custom flag I raised on SO recently: This appears to be ChatGPT-generated. The GPT detector gave it X% as how likely it was to be AI-generated. There might be a better flag, but I typically have a bit of info to give (the detector result and if the user has other ChatGPT answers), so I always use a custom flag. A mod can confirm, but I'd guess custom-flags are the best way
    – cocomac
    Commented Jan 19, 2023 at 17:44
  • @cocomac by "custom flag" do you mean the bottom-most option, "in need of moderator intervention", combined with a textarea? If yes, that's the basis for my question: I was looking for something that did not need additional mod attention. I was looking for something more automated, e.g. how I believe the spam flag works (auto deletion after like 5 votes).
    – Levente
    Commented Jan 19, 2023 at 17:48
  • @cocomac yes that's a custom mod flag. Moderators give attention to lots of things every day, including the non-custom flags. Do not feel bad about writing custom text if it is a legitimate reason. You should probably not strictly rely on the GPT checker though, as it has been known to flag human-written text sometimes. You can corroborate it with things like the user writing many answers in a short time, writing answers with info that looks okay at first glance but is actually nonsense, consistently failing the checker, etc. Commented Feb 1, 2023 at 18:31
  • I am still trying to figure out if @cocomac 's first comment above has been answered? Can AI be used for improving questions and comments or perhaps simplifying answers? Commented Feb 4, 2023 at 4:35
  • 3
    @C.S.Cameron The issue on AU has never been people using technology like ChatGPT wisely, it has been people using ChatGPT to game the Rep system, to produce empty and many times misleading answers in bulk, to easily create spam and flood AU with it. Hence a blanket and complete ban on its usage.
    – andrew.46 Mod
    Commented Feb 4, 2023 at 5:39
  • 1
    @andrew.46 So the answer to cocomac's comment is "No, there is a complete and total ban on ChatGPT and similar AI on AU. It can not be used in any way, not even to improve questions or comments or when editing". Is that correct? Commented Feb 4, 2023 at 7:40
  • 6
    @C.S.Cameron Well they are actually your words not mine 🙂. The first statement in the announcement pretty much covers it all: "There is now a permanent and complete ban on the use of ChatGPT and AI generated content on Ask Ubuntu." I cannot really add anything more to that...
    – andrew.46 Mod
    Commented Feb 4, 2023 at 7:53
  • 1
    @thomasrutter Sometimes there are tell-tale signs - people who play with ChatGPT a lot may notice. It writes relatively formally, for example. Complete sentences only; paragraphs starting with words like "First," and "However,". Another sign is if the answer is confident but subtly wrong (requires you to have actual knowledge to judge the content). Another hint is a user who posts many answers on completely unrelated topics in succession. Commented Mar 1, 2023 at 10:19
  • 1
    Here you go: Title: GPT-generated content Description: This question/answer appears to be generated by a bot or appears to be generated from ChatGPT (END) (P.S feel free to improve it when giving it to the mods)
    – Rishon_JR
    Commented Mar 10, 2023 at 5:51
  • 1
    @user253751 Relatively formal writing with paragraphs starting with "First," or "However,"? That sounds like me and I don't always use sentence fragments either. I'm on the autism spectrum, I aim to write as I speak, and I actually speak like that. Here's hoping people will focus enough on the other signifiers.
    – ssokolow
    Commented Jun 2, 2023 at 0:55

7 Answers 7

13

I am fine with this. To date, the times I have asked ChatGPT for code or linux help, it's been wrong. The whole point of AU and SE in general is learning and helping other. Perhaps the should start a "How to Ask ChatGPT" SE group.

3
  • 3
    My experience is more positive, I've had it help me with a lot of things , but I've always had to go over the code with an eagle eye to spot the crazy. It DOES do some absolutely berko things at times though, inventing entire imaginary libraries, or even trying to correct code that doesnt exist (Oh and for added comedy, dont even think about giving it URLs to read code at, it'll just completely hallucinate the contents and seem mystified when you tell it that its wrong. But in given limited scopes where you fully understand the problem ,its just tedious to DIY , it can be great.
    – Shayne
    Commented May 17, 2023 at 12:51
  • " I have asked ChatGPT for code or linux help, it's been wrong" how about using models specifically designed for code generation, such as GitHub Copilot? Commented May 18, 2023 at 17:26
  • ChatGPT is extremely good at generating both Python and Bash code. Of course the user must check what has been generated and verify it before running it or posting it on the web. Commented Sep 7, 2023 at 13:44
4

Basically I concur. ChatGPT has been proven to be unreliable, thus a ban is justified. However I've tested some code writing and was surprised that it worked.

I just wonder how we want to figure out if ChatGPT was the author?

3

This is sensible. Users still have the option to ask ChatGPT or another LLM themselves if they wish to. However, on a platform where every answer has a person's name attached to it, users should be able to safely assume that it was written by that person.

Current LLMs are often not correct in the information they provide, but format it in a highly convincing manner. When a user gets output from a LLM directly, they may know to scrutinize it accordingly. However, when this output masquerades as something produced by a person, the user doesn't have this opportunity.

2

AI is dangerous and this rule helps bring it to light. If everyone comes to depend on AI solutions, then the reason to learn and learning would cease. It would cause mankind to give up independence, control and ingenuity; going against everything open source stands for. Dependence upon AI would leave the mind a fruitless waste land, and unable to accomplish the simplest of task.

0

I can understand the reasoning behind this ban, particularly as many AI language models are still continually being developed and tend to make (often unnoticeable) mistakes that can drastically affect the usefulness of the information (or, in some cases, such models lie outright).

However, how does one go about proving their innocence, if a GPT detector falsely interprets a human-written answer to be AI-generated? No detection tool is perfect; I've seen human-written content be recognized as AI generated before (and the same vice versa, AI-generated content being identified as written by a human) - if a user has their answer deleted/reputation deducted, what can they do to appeal (and hopefully overturn) the decision?

Also, does this ban apply even to answers that only contain a small piece of AI-generated content, and if so, does providing attribution (and verifying the correctness of the content) make any difference? For example, if a user uses ChatGPT to generate a Linux shell command, which they have included in their (more detailed) answer to a question - is this use of AI permitted?

4
  • 1
    "if a user uses ChatGPT to generate a Linux shell command, which they have included in their (more detailed) answer to a question - is this use of AI permitted?" Without personally verifying that it works as advertised, no, it's not permitted: the reason is detailed in your first paragraph. However, if you personally verify on your machine that that command 1.) works as promised, and 2.) does not wreak havoc on your machine either, then you are just sharing a command that you had learned about recently, and you can vouch for its validity and safety.
    – Levente
    Commented May 18 at 19:51
  • 1
    We are here to help each other, reliably and responsibly.
    – Levente
    Commented May 18 at 19:51
  • 1
    On a further thought: if you are relying on ChatGPT to generate a command that you had not known about before, chances are that you have no deep experience with that command. There may well be additional considerations regarding some commands, and ChatGPT is certainly not guaranteed to inform you about such potential additional considerations. There is a risk that you end up propagating information in a confident tone of voice that you are not in the position of truly vouching for. Again: we are not here to merely produce answers. We are here to offer reliable help.
    – Levente
    Commented May 18 at 19:59
  • The AI detection tools are no longer used to detect AI generated material. Instead a set of heuristics are used to detect the material: meta.stackexchange.com/q/391990/662263. For the second part of your answer: I prefer not to delve into edge cases and simply to state that the use of AI tools on Ask Ubuntu is currently banned.
    – andrew.46 Mod
    Commented May 19 at 7:30
-9

As of February 2, 2023, there is a permanent and complete ban on the use of ChatGPT and AI generated content on Ask Ubuntu.

I disagree: we shouldn't blindly discriminate against AI. If some AI model provides a correct answer, then it shouldn't be deleted, provided it has the proper attributions. No need to waste human time if some AI can do it. However, if an SE user tends to post incorrect answers, then they should be banned. Ban users, not the tools.

11
  • 15
    That sounds nice, but it doesn't really work in practice. While AI models sometimes do produce a correct answer, they often produce answers that sound right but are totally wrong, and most reviewers can't easily verify answers to ensure they are correct. Also, some users post tons of AI-generated answers which is annoying and problematic
    – cocomac
    Commented May 18, 2023 at 18:09
  • @cocomac if an SE user tends to post incorrect answers, then they should be banned. Ban users, not the tools. Commented May 18, 2023 at 20:14
  • 7
    Well, we are not blindly banning AI content. We are actually, in a very focused way, trying to avoid anything like the tsunami of AI generated non-answers that is still afflicting Stack Overflow. I personally think that AI is a fantastic technology but its usage on Stack Exchange has always been very, very problematical.
    – andrew.46 Mod
    Commented May 19, 2023 at 1:29
  • 5
    When the tools are routinely wrong but look correct (or at least highly plausible), they're just not useful. The only way this tool is useful is when a subject matter expert carefully reviews the results, in which case they may as well have written the answer themselves. At this stage it's just easier and better for everyone if AI is banned outright. (Not to mention, if people want an AI-generated answer, they can get it themselves. They come to AU/SE for quality human-written answers.)
    – Clonkex
    Commented May 19, 2023 at 4:41
  • @Clonkex "The only way this tool is useful is when a subject matter expert carefully reviews the results" mostly true though some LLMs do provide attributions to some extent such as Bing Chat, where a nonexpert can review the links to check whether they support the LLM output. "Not to mention, if people want an AI-generated answer, they can get it themselves. They come to AU/SE for quality human-written answers.)" I disagree, they come for correct answers, regardless of who or what wrote it. And they can't directly use an LLM for that currently due to the inaccuracies of existing LLMs. Commented May 19, 2023 at 5:43
  • 1
    @andrew.46 "the tsunami of AI generated non-answers that is still afflicting Stack Overflow" sounds like the LLM ban didn't prevent the tsunami then? The ban doesn't prevent the posts anyway. It just means that instead of judging the correctness of the answers, one assesses the likelihood of whether the answer was written with a language model. Commented May 19, 2023 at 5:48
  • @FranckDernoncourt A bit hard to judge whether the ban has had an effect in SO, the only real way would be to stop the ban and see if usage took off. And that would be a foolish gamble. And as for correctness of the AI generated answer: for the most part they do not actually answer the question, certainly the many, many cases I have dealt with on AU have this in common.
    – andrew.46 Mod
    Commented May 19, 2023 at 7:48
  • 3
    @FranckDernoncourt Well... there are a ton of SO answers that get posted and are AI-generated. And... it would take a lot of reviewer + mod time to test them for correctness. It is way easier/faster to say "They are banned, and if we catch you (I can't share specifics there, but there are good tools we have to find them), then the mods can give you a suspension" (checking for AI-content is faster and more scalable than testing if it is correct).
    – cocomac
    Commented May 19, 2023 at 18:33
  • @cocomac that's ok, I know the specifics meta.stackexchange.com/a/384625/178179 point taken but I disagree with letting bad users making SE less efficient for good users. Commented May 19, 2023 at 21:11
  • 3
    @FranckDernoncourt Judging the correctness of the answers is only possible as an SME. That's the problem. When real people post incorrect answers, they almost always look wrong or low-effort. The problem with AI-generated (or LLM-generated, as you seem to prefer) answers is that they're often indistinguishable from human-generated answers while still being actually wrong.
    – Clonkex
    Commented May 19, 2023 at 23:06
  • Actually looking back at this seems like a good idea, but instead of immediately banning them, give them 3 warnings before banning them.
    – Rishon_JR
    Commented Sep 11, 2023 at 11:09
-20

If you can’t beat them, join them. ChatGPT can give quicker and better organized answers. If you want it you get. 1 minute no one else needs to be involved. No one gets upset at you for not following a strict set of rules. Espically for newbies, people will prefer ChatGPT. If we allow GPT content that will allow us to have far content then we could otherwise and attract more users.

8
  • 24
    I respectfully disagree. If a user wants a ChatGPT answer, then they can just ask ChatGPT directly. Also, it would result in people posting lots of AI-content in hopes of easy rep. Lastly, ChatGPT can be very confident but wrong, so if a human will manually verify the entirety of a response from ChatGPT, then they should just write an answer themself
    – cocomac
    Commented Apr 11, 2023 at 15:40
  • Those are all true of humans except the first. As to the first, that is true. And therefore stack exchange will be less used @cocomac Commented Apr 11, 2023 at 20:31
  • 17
    "ChatGPT can give quicker and better organized answers." Than what? Some of the users? Sure. But it's often completely wrong, which is kind of a bad thing on a Q&A site. Commented Apr 11, 2023 at 21:58
  • 2
    And humans are never wrong. Right @OrganicMarble Commented Apr 11, 2023 at 22:12
  • 13
    What you describe has absolutely no correlation to the ChatGPT material that we have seen on Ask Ubuntu. We have seen instead users who abuse ChatGPT to post multiple answers across AU and often across SE. These answers are usually a poor match to the original question, are often incorrect and have rarely if ever been tested by the person posting the answer. This is why ChatGPT has been banned on Ask Ubuntu and more widely in many sites of Stack Exchange.
    – andrew.46 Mod
    Commented Apr 12, 2023 at 1:14
  • By your logic human answers should be banned as well. We see answer that are mass posted. Sometimes they are wrong, dont answer the question, and dont work. However, we still deal with the occasional bit of nonsense. Dont throww out the baby with the bath water @andrew.46 Commented Apr 14, 2023 at 10:35
  • 6
    @Starshipisgoforlaunch Actually I would contest the idea that you have given: "We see answer that are mass posted." This is not seen on AU in my recollection, certainly not with legitimate answers. However one pattern of AI generated answers is something like 5 or six complex answers posted within the space of an hour, and often this pattern is then repeated across multiple areas of Stack Exchange.
    – andrew.46 Mod
    Commented Apr 15, 2023 at 7:43
  • 11
    the outstanding benefit of Stack Exchange sites is the relatively high signal-to-noise ratio. Which is a result of its design, moderation and community systems. Increasing the noise is a sure way to overwhelm those three systems, either eventually or all at once. It's not like SE is in a competition to answer the most questions. It's true value, for me at least, is the quality of the answers and the context provided by the comments.
    – BISI
    Commented May 13, 2023 at 18:01

You must log in to answer this question.

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