Timeline for Policy: Generative AI (e.g., ChatGPT) is banned
Current License: CC BY-SA 4.0
43 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jun 19, 2024 at 18:40 | comment | added | kaiserlautern | Go out, LLMs! Go build a Stack Overflow of your own and leave ours alone! | |
| May 14, 2024 at 7:54 | comment | added | aggregate1166877 | @AbdulAzizBarkat That's makes a lot more sense. Thank you for the reply. | |
| May 14, 2024 at 7:46 | comment | added | Abdul Aziz Barkat | @aggregate1166877 the current rate limit is actually 90 minutes, it was increased due to users posting AI generated content. | |
| May 14, 2024 at 7:42 | comment | added | aggregate1166877 | "Users with < 125 rep must wait 3 minutes between answers" - You can type answers in under three minutes? Bruh, am I dumb or something, it takes me years to type an answer that's actually useful. | |
| Aug 9, 2023 at 9:36 | comment | added | MbPCM | why dont stackoverflow should automatically add answer provided by the chatGPT, with a comment that - this answer was generated by chatGPT, for correction in question or further assistance , click here to open that question in chatGPT website." then the reader will decide if it want to test this answer or other's answers. | |
| Jun 22, 2023 at 6:07 | comment | added | Eyad Bereh | "Writing good answers takes time" ... The issue is that nobody on this site has given me a good answer for years (if at all). As soon as the generative AI gets better and starts producing more correct and accurate responses people will turn their backs on StackOverflow and nobody will use it anymore. | |
| Jun 17, 2023 at 9:30 | comment | added | Someone | I don't think it should be '1 per hour' but maybe '3 per 3 hours'. Similar questions can have varying answers but the same research. Once per hour would be downgrading, on the other hand, increasing the interval absolutely but not relatively would be an interesting option to look into. | |
| May 9, 2023 at 2:48 | comment | added | Faron | This is opening a can of worms. This kind of discussion is already inevitable. ChatGPT is now on rise, period … whether you like it not. More people are now accessing ChatGPT, and there is nothing to stop them. The question that is remains to be seen: do we want waste our time trying to stomp out ChatGPT when there is a huge rise of users for last 7 months? Solution: Regulate it to an acceptable level. We now have tools that can detect ChatGPT choices of word in all of their output. We just need to be responsible for outputs to be accurately correct. We only are ones that can do that. | |
| Apr 18, 2023 at 11:39 | comment | added | Xavi Montero | Even, to make it more attractive, if you write say 10 answers in a row, instead of rejecting the posting, they could be accepted and "queued" so they are "released" at a reasonable rate. For example users between 1 and 100 rep could have a rate-limit of 1 per week. No matter if they do 10000000000 posts with GPT only 4 will be released in a month. If a "real user" generates 100 great human content-verified answers and also is new to the site, the 100 anwers will progressively be pushed forward over 1 year. The quality of the first answers will take part in the equation for the allowed rate. | |
| Apr 18, 2023 at 11:32 | comment | added | Xavi Montero | I even take not minutes but hours to write a good answer! I'd be happy if I was limited to an AVERAGE of 1 per day. But sometimes I'd like to post 2 or even 3 per day. Consider limiting on the "week-average" or "month-average". I don't know where are the limits, but the Stack Exchange staff will have access to usage statistics. Take some 5k-rep users, and some 15k-rep users and check their "yearly-rate" over the last 10 years. That gives an idea on where sane limits "on average" could be set. | |
| Mar 27, 2023 at 9:34 | comment | added | Karl Knechtel | Questions that are properly clear and focused that can be readily answered en masse, for a mature technology like JavaScript, really ought to be duplicates already. | |
| Feb 20, 2023 at 19:48 | comment | added | Travis J | If this were to take place, with a throttle, then there is a higher likelihood that this type of widespread behavior would go unnoticed. Further, it would infringe on legitimate use. There is good intention here, but the ends do not justify the means. | |
| Feb 8, 2023 at 16:46 | comment | added | jfs | downvotes may be used to affect the rate drastically. | |
| Feb 7, 2023 at 10:42 | comment | added | M.Stramm | I would like to mention that the captchas are pretty useless, because these bot-users likely do not create answering bots. The policy should not target fully automated answering bots, it should target users who are manually copying questions to CPGT, then manually copying the answer to SO. Those users are not stopped by captchas, they are merely slowed. Also they are not deterred by rate limits or captchas, since they still have to put in very low effort to farm reputation compared to legitimate users. | |
| Dec 20, 2022 at 18:50 | comment | added | Lyndon Gingerich | If undesirable answers are being generated automatically, then the scripter can just adjust his program to include a timeout adhering to the limits, then let the program run all night. Scripters will have an easier time than honest answerers adhering to rate limits. | |
| Dec 20, 2022 at 17:05 | comment | added | JL Peyret | At this point the limit of writing 5+ good answers in under an hour on a subject you are expert in isn't someone's skill, it's finding good questions to answer. Python questions asking really basic stuff are a dime a dozen. Many noob questions however are either dupes, get closed for lack of research or other flaws. Or - even more common - answers never get accepted. So, again, more issues due to the question quality than any real difficulty farming low hanging fruit if someone is so inclined. The 30min rate limit on <125 rep users makes total sense however. | |
| Dec 15, 2022 at 1:34 | comment | added | mirabilos | Good choice, also because text generated by ML (so-called “AI”) programs is also typically illegal in virtually all cases because it’s a deterministically derived work of its inputs. | |
| Dec 9, 2022 at 6:24 | comment | added | Markuzy | Ironically, I think you need an AI-assisted modding tool to flag users with certain characteristics of answering rate, the types of answers given, and then these will be targeted for manual review. Someone will eventually write a bot that integrate with ChatGPT, and input questions from Stackoverflow, retrieve and cache the answer, and then output to Stackoverflow every 30 mins... | |
| Dec 9, 2022 at 1:36 | comment | added | magnump0 | @NineBerry the problem is not in frequency, but in quality. I'd vote for high penalty for users posting answers that do not work. | |
| Dec 8, 2022 at 19:47 | comment | added | starball Mod | Update: in the staff announcement on their ChatGPT policy, for users with <125rep, their answering rate-limit is now 30 minutes (was 3). This change is also explained in the Help Center. | |
| Dec 8, 2022 at 8:06 | comment | added | ChrisW |
@Dharman I am really doubtful that someone can find 5 good questions and write a good answer to each one in less than an hour.. Jon Skeet could, for sure. Seriously though I don't think I've stopped to "test" an answer before posting it -- if I posted it's usually because I "knew" the answer after reading the question, and so I just write it.
|
|
| Dec 7, 2022 at 21:47 | comment | added | David C. Rankin | A real answer, even a simple one, should take no less than the 5 minute edit grace period to produce before you press the "Answer" button again on another question. A well formatted and referenced answer can take 10 times that to produce. | |
| Dec 6, 2022 at 11:03 | comment | added | CrandellWS | extreme temporary measure are in order and even if it restricts me temporarily I am 100% for it, heck I am 1000% for it ... Stack must withstand the AI revolution for the good of all | |
| Dec 6, 2022 at 5:42 | comment | added | polkovnikov.ph |
Clipboard API supports adding a custom mimetype to data in clipboard. If OpenAI just added a text/x-chatgpt for copied text, at least other resources would have a protection against foolest of fools that copypaste directly from their website.
|
|
| Dec 5, 2022 at 22:53 | comment | added | mcint | Answers under a certain length should be keep old limits. | |
| Dec 5, 2022 at 22:02 | history | edited | cottontail | CC BY-SA 4.0 |
added 4 characters in body
|
| Dec 5, 2022 at 18:59 | comment | added | The_spider | @Cerbrus It doesn't solve the problem, but its better then that they can post hundreds of bad answers a day. | |
| Dec 5, 2022 at 18:56 | comment | added | The_spider | I do think that higher limits for posting answers are a good thing. However, I think one hour is just a bit too long. I would rather go for a quarter to half an hour. | |
| Dec 5, 2022 at 18:47 | comment | added | 9072997 | I like this idea, but I would implement it as a "leaky bucket" rate limit. Normal users have to sleep, and thus would be less affected. | |
| Dec 5, 2022 at 17:12 | history | edited | NineBerry | CC BY-SA 4.0 |
edited body
|
| Dec 5, 2022 at 15:33 | comment | added | camille | I'd love to see how this plays out for the folks who post the same answers to bad questions 20 times a day without using bots | |
| Dec 5, 2022 at 12:53 | history | edited | NineBerry | CC BY-SA 4.0 |
added 1 character in body
|
| Dec 5, 2022 at 12:53 | comment | added | Cerbrus | @JiříBaum So a user just dumps, say, 10 low quality answers on SE every day... Rate limits don't solve this problem. | |
| Dec 5, 2022 at 12:52 | comment | added | Dalija Prasnikar Mod | @Dharman As an expert I can easily write several good and elaborate answers in an hour. If I have an hour now, that does not mean I will be free to write answers in a hour. Also if I save answer for later, question might already be answered by adequate answer and I don't like posting duplicate answers even if mine might be a better one, unless it is exceptionally better. | |
| Dec 5, 2022 at 12:50 | comment | added | Jiří Baum | A per-day limit (or other time window) might be the better option; that would allow a user to post several answers in one session while still limiting the overall rate. | |
| Dec 5, 2022 at 12:38 | comment | added | Cerbrus | Where do you think I got my rep? Note that I didn't say "good", I said "decent". My point is that we shouldn't be punishing honest users for the abuse from a few. | |
| Dec 5, 2022 at 12:38 | comment | added | Dharman Mod | @Cerbrus I am really doubtful that someone can find 5 good questions and write a good answer to each one in less than an hour.. | |
| Dec 5, 2022 at 12:38 | history | edited | NineBerry | CC BY-SA 4.0 |
added 23 characters in body
|
| Dec 5, 2022 at 12:37 | comment | added | Cerbrus | @Dharman browse the JavaScript tag. There's plenty of decent questions there that you can answer. Why shouldn't someone new to SO be allowed to answer 5 questions in an hour, if the answers are correct, and reasonably explained? | |
| Dec 5, 2022 at 12:36 | history | edited | NineBerry | CC BY-SA 4.0 |
added 25 characters in body
|
| Dec 5, 2022 at 12:36 | comment | added | Dharman Mod | I am fully on board with this idea. If you really have two good answers to post then waiting an hour is not an issue. I think posting an answer quicker than 1 per hour (regardless of rep level) is not good for the site. It takes time to search for a duplicate, test the code, write proper explanation and so on. | |
| Dec 5, 2022 at 12:34 | comment | added | Cerbrus | Rate limits don't force people to verify their content, it just makes them wait. It's also quite unfair to punish correct usage for the abuse from a few lazy bot users. | |
| Dec 5, 2022 at 12:27 | history | answered | NineBerry | CC BY-SA 4.0 |