Timeline for Policy: Generative AI (e.g., ChatGPT) is banned
Current License: CC BY-SA 4.0
53 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Apr 12, 2023 at 20:42 | comment | added | Ryan M Mod | @JamieB It is unsurprising that ChatGPT is able to effectively answer questions that are likely in its training data, as large swathes of the internet before a certain date are. | |
| S Apr 12, 2023 at 20:33 | history | notice added | Ryan MMod | Comments only | |
| S Apr 12, 2023 at 20:33 | history | locked | Ryan MMod | ||
| Apr 9, 2023 at 15:21 | comment | added | Donald Duck is with Ukraine | I just tried asking Bing's AI chatbot "why should stack overflow allow ai generated answers?" and it cited this answer as a source that AI-generated answers could potentially provide more accurate and detailed answers. | |
| Mar 22, 2023 at 14:09 | comment | added | anotherDev | This answer is meta on a meta level. | |
| Feb 14, 2023 at 22:20 | comment | added | Phil | @LemonDrop - your questions were essentially "why X" and "why not X", not "should X", so it makes sense that you got arguments for and against, rather than a weighing in for or against. | |
| Feb 2, 2023 at 11:01 | comment | added | S. Dre | I find highly ironic that ChatGPT has the right and most upvoted answer to this post about not allowing ChatGPT answers on SO partially because of it's low rate of correct answers. | |
| Dec 30, 2022 at 12:58 | comment | added | nicomp | "AI-generated answers may not always be accurate or helpful," -- just like my answers! | |
| Dec 30, 2022 at 10:07 | comment | added | Lu Kas | "It is very sensitive to presumptions in the query itself". Its answers feel very much like a university student trying to answer an open ended exam question like "Explain why ...", mentioning all the possible reasons they can think of, adding as much fluff as possible, without any critical thought, and without really knowing the real answer. | |
| Dec 27, 2022 at 5:31 | comment | added | Thiyagu | It answers to uphold confirmation bias | |
| Dec 18, 2022 at 5:43 | comment | added | JamieB | Anecdotal: I asked ChatGPT the question from stackoverflow.com/questions/9229601/what-is-in-c-code and it actually gave a very nice answer. If the asker had asked chatGPT, he wouldn't have needed to ask us. It does have its uses. Sometimes, though, it "misses the point" of a question or just yields an objectively incorrect answer. | |
| Dec 17, 2022 at 4:13 | comment | added | Eric Haynes | And important distinction is to realize that StackOverflow will NEVER be an appropriate place to post AI generated answers. What would be the point? If the AI ever reaches the point of giving good answers, you would just stop visiting StackOverflow. It could be renamed, "shit-the-bot-said-when-it-was-dumber.com". Of course, if it gets to that point, you'll likely be asking it "what career options are there for a former software engineer". | |
| Dec 14, 2022 at 15:12 | comment | added | osiris | @JohnBeale, "Probably no human has read as much source material as the AI, while being mostly unable to logically reason about it or check for correctness." -> You'd be surprised... | |
| Dec 12, 2022 at 7:33 | comment | added | NoDataDumpNoContribution | @radj At some point I had the idea of always writing two answers to every meta question, one arguing for and one against the question because I can and to show that the real value of an answer is not the content alone but the content with a score. Votes are incredibly important, while answerers are not needed to believe in their cause, especially for opinionated questions like they appear on meta. But then I'm maybe a bit too much of a machine. :) | |
| Dec 11, 2022 at 22:43 | comment | added | radj307 | @Trilarion You can instruct ChatGPT to have an opinion, which works really well for a lot of things, such as this project that instructs ChatGPT to pretend to be a hypothetical safety engineer to evaluate whether prompts should be sent to itself or not. I played a text-adventure game that I told it to automatically generate for me based on whatever prompts I wrote for it. At first it refused to fill in any details that I didn't specify, but it always gave me recommendations so I told it to use those automatically and it worked! | |
| Dec 11, 2022 at 22:38 | comment | added | radj307 | It can also generate surprisingly complex code (especially Python code) that usually works! | |
| Dec 11, 2022 at 19:35 | comment | added | René Roth | AI might never become perfect. But are human contributors? The rambling style of GPT generated texts just reeks of human bullshitting you see all too often on the internet. No, it's not perfect, but it absolutely passes as user generated content. | |
| Dec 9, 2022 at 9:14 | comment | added | NoDataDumpNoContribution | @StoneLam "It only gives out answers you want to hear. Not the one which is correct" In this case there isn't really a correct answer. You can argue for one side or for another and what surprises me how well argued both answers are. If anything one could accuse the bot here that it obviously doesn't have an opinion of its own. But otherwise the quality of the answers given is great, a contrast to the reported poor quality of answers to programming questions in this announcement here. | |
| Dec 8, 2022 at 23:48 | comment | added | Stephen | I call BS on this one: "For one, AI-generated answers could potentially provide more accurate and detailed answers to users' questions. This could be especially useful for complex technical questions that may be difficult for human experts to answer." ChatGPT's answers are not particularly detailed, they mostly fluffy answers. | |
| Dec 8, 2022 at 20:42 | comment | added | Chris | Reads like a congressional hearing. Lots of words with near zero content. | |
| Dec 8, 2022 at 9:44 | comment | added | John Hunt | I thought this answer was perfectly valid. I also think there's a lot of concern about AI 'stealing our jobs' which is just stupid. Personally I think it's a good tool. I also agree it can produce misleading/wrong answers to questions so the ban is a good idea although it's going to be difficult to implement the ban. SO is the thing that will suffer most from ChatGPT I believe, which personally I don't care that much about. ChatGPT will improve over time and make SO (fairly) redundant I believe. | |
| Dec 8, 2022 at 2:23 | comment | added | Shayne | Keeping in mind the philosophical oddness of applying a term like "stupid" or "clever" to an algorithm, I actually think its remarkably clever. Sure its wrong, but the fact it reads so well and could easily fool a human into believing it was another human (at least until one has a bit of a think about it), tells me these things are getting very close to something we might have to eventually conceed is in fact "general" AI. Unfortunately for us, the danger from AI isn't "AI does a stupid thing", its "Stupid humans using AI to do stupid things." | |
| Dec 7, 2022 at 18:50 | comment | added | Samathingamajig |
@CrandellWS I love this, not only because "had the AI allegedly responding", but because they put 'Sam' in single quotation marks. Like, they both shortened my name and put it in quotes (I put that my name is Sam in my profile) as if that was allegedly
|
|
| Dec 7, 2022 at 18:28 | comment | added | hedgedandlevered | the top answer being 90% AI generated content is one of the most ironic things I've ever seen. Seriously great job. Bonus points for this being on a "meta" site | |
| Dec 7, 2022 at 18:27 | comment | added | CrandellWS | @Samathingamajig you name made the news cause of this bleepingcomputer.com/news/technology/… | |
| Dec 7, 2022 at 16:54 | comment | added | BJS3D | After skimming some of the comments to this, I've found a bit of irony in @ShadowTheKidWizard's take on the matter. AI's inability to effectively answer questions is not a shortcoming of AI, it's a shortcoming of its creators. In effect, whenever someone concludes that "AI is this" or that "AI won't ever achieve that", they're effectively saying that humans are ill-equipped to create an AI capable of more than being novelty programs. The failure isn't with the AI. Rage on, though. :D | |
| Dec 7, 2022 at 9:22 | comment | added | luator | @MichaelAnderson I asked ChatGPT why SO should ban answers written by humans. The reponse: imgur.com/s2UeGCr I especially love the point about plagiarism :D | |
| Dec 6, 2022 at 20:26 | comment | added | John Beale | It produces grammatically well-polished answers that for many (most?) non-trivial questions are simply wrong. Yes, plenty of wrong human answers, but usually with more hints the writer is non-expert. Probably no human has read as much source material as the AI, while being mostly unable to logically reason about it or check for correctness. | |
| Dec 6, 2022 at 18:05 | comment | added | Martin Braun | Interestingly though, it will only include counter arguments to the question when being asked why it should be used on StackOverflow. So it will not include reasoning to include it on StackOverflow when asking why not to use it on StackOverflow. | |
| Dec 6, 2022 at 16:53 | comment | added | Ray | @ShadowTheKidWizard It's not stupid, exactly; it performs it the task it was trained for very well. It's just that said task is "generate grammatically correct, idiomatic English text", not "provide factually correct answers". The error is in attempting to use it for a task it was not designed for. (Which in turn is caused by the error of overhyping these models.) | |
| Dec 6, 2022 at 13:14 | comment | added | Peter - Reinstate Monica | "AI-generated answers could potentially provide more accurate and detailed answers to users' questions. This could be especially useful for complex technical questions that may be difficult for human experts to answer": Classic Danning-Kruger. I mean, sure, it's doing what it is designed to do: Project confidence without having the slightest clue. But still. | |
| Dec 6, 2022 at 9:56 | comment | added | LunaticNeko | One of the biggest matters here is that whenever asked if something is a good idea or a decision should be made, OpenGPT seems to always end every response with "whatever I said, both A and B have their merits so you should use discretion" kind of answer. I think it's one of the giveaways that it's probably not a human, because people on the Internet these days don't seem to always be neutral like that. | |
| Dec 6, 2022 at 2:08 | comment | added | StoneLam | It only gives out answers you want to hear. Not the one which is correct. | |
| Dec 5, 2022 at 17:42 | comment | added | Fabio says Reinstate Monica | Holy cow, AIs can already write answers of this quality? Please tell me I'm not the only one who's shocked! I wouldn't have guessed in a century that these are AI-generated! | |
| Dec 5, 2022 at 13:16 | comment | added | Cerbrus | @nmeln Mission Friggin' Accomplished! | |
| Dec 5, 2022 at 13:08 | comment | added | nmeln | @Dalija Prasnikar Ban policies are not going to scale. What will happen if OpenAI trains a much better ChatGPT that generates actually factually correct answers for tech questions? What if StabilityAI does the same? Unless companies also provide free and open source tools to verify that answer is AI-generated, human moderators are going to have hard time keeping up with these advances | |
| Dec 5, 2022 at 12:53 | comment | added | NoDataDumpNoContribution | Great answers. Can I copy them? What license are they under? | |
| Dec 5, 2022 at 12:41 | comment | added | Dalija Prasnikar Mod | We should keep Stack Overflow clean for questions (problems) and answers that require people to answer correctly. | |
| Dec 5, 2022 at 12:40 | comment | added | Dalija Prasnikar Mod | @nmeln To what end? If the users need to mark the answer is AI generated, then it is very likely such answers will be instantly downvoted because most of those answers are wrong and users posting such answers only make moderating here harder. But let's say that user is posting such answer after verifying it is correct. If the user has knowledge to know answer is correct then they have knowledge to write it themselves. Even if the AI would be generating correct answers, then asking questions here that could be answered by AI would be also a waste of everyone's time. | |
| Dec 5, 2022 at 12:34 | comment | added | nmeln | @Dalija Prasnikar There is another solution, either ask users to mark if their answer is AI-generated (difficult because some users are in it for reputation points), or SO can integrate with chatGPT to provide "best effort" answers to unanswered questions. | |
| Dec 5, 2022 at 10:16 | comment | added | Cerbrus | "and contains about the same amount of thought (i.e., none)." Well that made my day! xD | |
| Dec 5, 2022 at 9:53 | comment | added | GOTO 0 | Have to agree with @KarlKnechtel. This is not worse in any way than most human-written answers I read on meta. | |
| Dec 5, 2022 at 9:47 | comment | added | Karl Knechtel | This is, in terms of English technical proficiency, better than a lot of high-school and university-level writing I've seen; and contains about the same amount of thought (i.e., none). | |
| Dec 5, 2022 at 8:46 | comment | added | l4mpi | "AI-generated answers could potentially provide more accurate and detailed answers to users' questions. This could be especially useful for complex technical questions that may be difficult for human experts to answer" - That's a good one. It might not be great for generating answers based on facts instead of bullshit, but it seems pretty decent as a joke generator. | |
| Dec 5, 2022 at 8:22 | comment | added | Dalija Prasnikar Mod | @MichaelAnderson the difference is that poor human answers are also rather recognizable. They usually lack explanations or are poorly written. AI generated answers look like genuine good answers and only fall apart when you try to apply them as they are mostly incorrect. Also people alone cannot possibly generate such amount of incorrect answers like AI can. There are SO users active for years that previously produced only few answers now posting over 50 in less than a day. The amount of AI generated answers could suffocate SO if everyone starts doing it. | |
| Dec 5, 2022 at 7:45 | history | edited | Jean-François FabreMod | CC BY-SA 4.0 |
deleted 52 characters in body
|
| Dec 5, 2022 at 7:43 | comment | added | Michael Anderson | My favourite thought question is "to what extent could these reasons apply to banning human answers. Humans are pretty guilty of "answers may not always be accurate or helpful, and there may be concerns about the potential for bias... " | |
| Dec 5, 2022 at 6:51 | comment | added | Lemon Drop | @Samathingamajig Sure but it illustrates why it's not very useful. It is very sensitive to presumptions in the query itself (like if using it in SO answers is a good thing or not). It's not doing any actual thinking nor does it have consistent reasoning so it easily will fall into telling you what you want to hear when giving loaded questions like this. To me that just makes it about as useful as a Google search just with a much more well-spoken presentation given its understanding of English at least. | |
| Dec 5, 2022 at 6:36 | comment | added | bombcar | Interesting that both answers bring up the inaccuracy. | |
| Dec 5, 2022 at 6:30 | history | edited | Samathingamajig | CC BY-SA 4.0 |
added 1064 characters in body
|
| Dec 5, 2022 at 6:24 | comment | added | Samathingamajig | well, @ShadowTheKidWizard, it does have the correct answer in this case... | |
| Dec 5, 2022 at 6:24 | comment | added | user447356 | it's perfect and proves how stupid this thing really is. | |
| Dec 5, 2022 at 6:18 | history | answered | Samathingamajig | CC BY-SA 4.0 |