Timeline for Policy: Generative AI (e.g., ChatGPT) is banned
Current License: CC BY-SA 4.0
58 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jan 12 at 21:07 | comment | added | Fattie | The fundamental issue is that Gen-AI is completely useless. If Gen-AI "worked" and AI "beings" could understand and answer programming questions (other than the most trivial - which is nothing more than a re-factoring of a google search) then the whole world would be using Gen-AI software and the whole question would be moot. | |
| Jan 12 at 19:53 | comment | added | Cerbrus | @ChatGPT it looks like you missed the entire reason this policy was put into place. LLM-generated answers were flooding the network, and there wasn't enough manpower to "focus on whether or not they are correct or useful". Their good/bad answer ratio wasn't anywhere near to what you'd want. We trust SO. We just (rightfully) don't trust LLMs. | |
| Jan 12 at 2:40 | comment | added | maxhodges | Instead of worrying if answers are AI generated or not how about if we just focus on whether or not they are correct or useful? Stack overflow is already built for that. We don’t need a special class of “hate crime” style rules which are redundant with existing rules and policies. No one cares if an answer was generated by carbon-based lifeforms or silicon; people care about solving problems. If you have any trust of SO, you’d let it work. It was designed to filter up good answers. | |
| Jan 3 at 18:20 | comment | added | Fattie | ? everyone has given up on generative AI for coding, it was a fad that passed. fierce-network.com/cloud/generative-ai-hype-dying | |
| Oct 17, 2023 at 13:25 | comment | added | Fattie | hi @ChatGPT , I realize you were just being funny, but your phrase "the rest of us" means "the set of beginner and/or mediocre programmers". I'm not in that set. As I've mentioned to you over and over and over, ATM chatGPT is good for very basic hobbyist-level programming questions. ATM it's useless for bleeding edge, or simply "difficult" questions. I could simply copy and paste here my whole reply to Dan (just above). The instant it is useful to me, I will use it constantly, 8 hours a day, don't know how often I can repeat this. ATM useless for bleeding edge or difficult questions. | |
| May 15, 2023 at 21:42 | comment | added | Fattie | (And I don't want chatgpt answers posted on here because there is ALREADY a staggering amount of poorly written, often just bad, answers to both trivial and more valuable questions , which really clutter stuff up.). | |
| May 15, 2023 at 21:41 | comment | added | Fattie | ... that's my take | |
| May 15, 2023 at 21:41 | comment | added | Fattie | chatGPT is totally useless other than for the simplest of questions. (and you can just instantly google those.) As soon as I heard about chatGPT I instantly went to a laptop and started trying to use it in my work. it has never helped me once, I report to you. The instant it is useful to me, I will use it constantly, 8 hours a day. Hence, Dan, "but robots are getting competent more quickly than we're accustomed to seeing" I just don't see it. You may well be right, even 'soon' - if so, the first person you'll hear it from is me! I hope that makes sense | |
| May 15, 2023 at 20:27 | comment | added | danh | I guess I'm in the minority about this, very possibly because I'm thinking about it wrong, but robots are getting competent more quickly than we're accustomed to seeing. It seems to me that a big wave is coming and that an operation like this one is either going to figure out how to surf it or get pummeled. | |
| May 15, 2023 at 20:12 | comment | added | user400654 | @danh yea, no, there's no value in critiquing AI generated answers. If one is capable of properly critiquing an AI generated answer, one is capable of producing a proper answer. | |
| May 15, 2023 at 20:07 | comment | added | danh | But @VLAZ, I'm not offering the idea to dissuade human knuckleheads from polluting the site - I don't know how to do that. What I'm suggesting is that the robot answers often are critique-able, edit-able into something useful (more often than with malfunctioning humans), and it's maybe a really good human learning process to do that work, and to review others doing that work, and to do it with no human feelings being hurt (which happens here, too often). | |
| May 15, 2023 at 20:06 | comment | added | danh | I think I understand what you're saying now, @VLAZ, since a robotic answer posted by the system and a robotic answer pasted by a human knucklehead will differ, the humans won't be dissuaded from posting. | |
| May 15, 2023 at 19:41 | comment | added | VLAZ | @danh because there is no "one" answer that ChatGPT can provide. Your entire suggestion hinges on providing one answer. Where ChatGPT can provide you with a wide range of answers. A lot of them wrong as in "not even in the same ballpark" where the only critique that can be added is "That's not at all how any of this works". Which isn't really that useful to learn from. The real reason to ban it is that it's too often dangerously misleading and wrong on even rather basic topics. And users are spouting that nonsense on the site without any regard other than getting points. | |
| May 15, 2023 at 19:36 | comment | added | danh | @VLAZ , sure, but why is their non-determinism a reason to ban vs. co-opt the bots? | |
| May 15, 2023 at 18:04 | comment | added | Fattie | @Thismatters very nicely put. The most striking innovation in QA forums is also an Achilles heel. As a matter of fact, for myself I deliberately give away my points as fast as I can (with bounties) for that very reason - part from anything else my real-life colleagues would consider it naff to have a "high score" on SO :) Actually you've reminded me it's time to find worthy bounties and "spend". | |
| May 15, 2023 at 17:05 | comment | added | VLAZ | @danh no, for many reasons. Key one - a gen AI can produce DIFFERENT ANSWERS FOR THE SAME INPUT. And I cannot understand why so many people somehow think it doesn't. | |
| May 15, 2023 at 16:58 | comment | added | danh | Another way to stop humans from posting robot code would be for the site to provide it proactively, marked as such. Treat it like an answer, up-vote, down-vote, etc. Well written critiques of bad code are sometimes more useful than good code. And from what I've seen, robotic answers are sometimes astonishing and probably on an exponential path toward better. I think SO needs to come to grips with this new stuff, looking to co-opt rather than do battle with it. However this works out, you have to admit it's fascinating! | |
| May 10, 2023 at 9:41 | comment | added | trevorcroft | To allow ChatGPT answers, one would need to enforce more "rigor" on the answers. For example, ChatGPT should provide evidence of how it tested the answer (ie: if it was a Visual Studio question it would have to provide screenshots), then stackoverflow would require AI reviewing the ChatGPT generated answers to verify the additional evidence, if the AI gatekeeper could "read" the screenshots and do it's own test to confirm the answer worked then the answer could be posted. So essentially pit one AI against another AI, the system could then be proven to be useless or useful. | |
| Feb 3, 2023 at 9:57 | history | edited | Wai Ha Lee | CC BY-SA 4.0 |
Consistency
|
| Feb 3, 2023 at 9:23 | history | edited | aljgom | CC BY-SA 4.0 |
Formatting. Got rid of the unnecessarily huge bold text
|
| Jan 18, 2023 at 17:46 | comment | added | Thismatters | @Fattie is hitting the nail on the head. The reputation system, and people's need to game it for their own supposed benefit (be it professional, academic, or just for the attainment of a massive e-peen) is the root of the issue with AI. The most striking innovation in QA forums is also an Achilles heel. | |
| Jan 15, 2023 at 6:12 | comment | added | Fattie | lol good one. chatGpt currently is useless (but excellent for easy, "facts available," questions) I'm a fan of the principal and I hope it improves and comes good one day (which would be amazing). | |
| Jan 15, 2023 at 5:12 | comment | added | maxhodges | talk about cognitive dissonance how can you state that GPT has never given you anything but "incoherent text" while at the same time you claim to be a fan o GPT? I can't have an honest discussion with a disingenuous person. @Fattie | |
| Jan 14, 2023 at 16:16 | comment | added | user19642323 | @MaxHodges "But I get perfect answers all the time." No, you don't, as you aptly demonstrated when you posted about ten of them to Japanese StackExchange. Pretty much all of them were plain wrong, so much so that it led to you getting a suspension from the site for a week. I struggle to see how you can maintain this doublethink that ChatGPT is perfect while you have clearly found through your own experimentation that it is not. | |
| Jan 14, 2023 at 15:13 | comment | added | Fattie | In a sense, MaxHodges, if you admire chatGpt, and, you believe it is the future and you wish chatGpt to be successful, the last thing you would want to see is chatGpt answers being pasted on to SO!!! For fans of chatGpt, I cannot see any value whatsoever in chatGpt answers being pasted in to SO. it's better in a business sense that everyone goes straight to chatGpt! cheers | |
| Jan 14, 2023 at 15:08 | comment | added | Fattie | @MaxHodges that is an example of a very straightforward beginner question (like asking the capital of a country, or spelling of a word). The challenging and useful questions on SO (which are usually about some new breaking change to an API) it does not have a clue. But this is irrelevant! Again! -> like you I'm a FAN of chatGPT and, indeed like you ,I guess that in the very long term chatGpt WILL REPLACE !!! SO. I'm telling you I agree with you that eventually chatGpt WILL REPLACE !!! SO. That is unrelated to whether it should be allowed to paste chatGpt answers on this site. | |
| Jan 14, 2023 at 2:51 | comment | added | maxhodges | @Fattie you wrote "every single question I've...asked chatGPT...in all cases it did not even produce coherent text." But I get perfect answers all the time. Example, "give me a regex for a six-character string that starts with "T" and then includes any capital letters or number" It returns "T[A-Z0-9]{5}$" If I ask on SO I get zero results. I could post it on SO and come back, but what if I can get that ChatGPT answer on SO and it becomes the accepted answer? SO grows more useful, and people get help faster. Not everyone is getting "only incoherent text" by any means. | |
| Jan 13, 2023 at 22:10 | comment | added | Fattie | BTW it's unimportant to the question on this QA, but the link you posted is a whacky publicity-stunt level event. I'm not saying ChatGPA won't "take over the world" but that reference is not a stepping stone. | |
| Jan 13, 2023 at 22:10 | comment | added | Fattie | @MaxHodges let's say that people asking questions, can indeed, get answers from "chatGpt". If so - (A) that's great for them and (B) as you say, SO will disappear as a business: nobody will use SO any more. The QA at hand here is about the unrelated, simple, issue of whether persons can generate text from chatGpt, and, post it here on SO as an answer. | |
| Jan 13, 2023 at 16:35 | comment | added | maxhodges | SO is dead. GPT was recently listed as an author in a research paper. twitter.com/kareem_carr/status/1613325029848092672 And the dead of Suffolk University Law School is adopting it in the classroom. And SO wants it banned lol | |
| Jan 4, 2023 at 12:23 | comment | added | Fattie | someone solved it. now, that answer appears on google (and hence obviously chatgpt, which is just google with dressing). this is completely unsurprising because tyhis is how it has always worked. i ask an obscure "new, difficult" programming question: ie, I google it. there's no answer, nobody has solved it yet. i put it on OS. someone figure it out. that answer, ie my own Q/A, is now the first result on google. this is how google has always worked (and now how chatgpt works, since it's google with dressing) | |
| Jan 4, 2023 at 12:21 | comment | added | Fattie | regarding your comment "SO owns a shit load of data and it should use it for its own benefit ..." ... regarding programming, chatGPT is getting ALMOST ALL of its information FROM SO. it's gotta be at the 95%+ level. exactly as, when you simply google a programming question, 95% of the time the answer comes from SO. it's no more complicated than that. for example, recently I asked a difficult question about a new swift api. there was no answer on SO (and hence: no answer on google, nor obviously on chatgpt which is just google with dressing ...) cont: | |
| Jan 4, 2023 at 12:18 | comment | added | Fattie | the issue at hand here is folks pasting chatgpt "answers" IN TO SO. if you wish to use chatGpt, whatever great. the issue at hand here is folks pasting chatgpt "answers" IN TO SO, which is incredibly annoying (to SO users, such as me). | |
| Jan 4, 2023 at 12:17 | comment | added | Fattie | @joesan if using chatGPT solves your problems, there's nothing else to say than "that's great". If chatGPT solved my problems I would use it continually, and you would never ever ever ever ever see me posting here again. however, as i stated above, it's useless for what i would call "actual difficult problems" (Random example looking at my recent questions, "How to add a placeholder text to UITextView, but, in all cases (even with changing OS) exactly match the text positioning?" the answer supplied is incredibly silly). .... | |
| Jan 3, 2023 at 19:43 | comment | added | user400654 | SO isn't a help desk, it isn't meant to be a place where you just dump your question and get an answer. Another service that offers that, whether or uses AI or not, would be better suited than SO for people looking for that kind of tool. | |
| Jan 3, 2023 at 19:40 | comment | added | joesan | I'm just trying to say that SO owns a shit load of data and it should use it for its own benefit so that it continues to stay relevant among the dev community. | |
| Jan 3, 2023 at 19:16 | comment | added | user400654 | @joesan there's a lot of people who use SO for a purpose it wasn't meant for. For those people, chatgpt may be just as good for their needs. SO doesn't need to serve that purpose to remain relevant. | |
| Jan 3, 2023 at 19:14 | comment | added | joesan | I trust the SO days are numbered if they do not adopt this AI generated answers. I love SO for what it has served, but this chatGPT is much better at solving my problems immediately. | |
| Jan 3, 2023 at 13:02 | comment | added | Fattie | I think it needs to bear in mind, the whole issue here is adding fluff. I couldn't care less if some total wanker, for some reason, wants to get " ' points ' " on some prosumption web site owned by a Dutch media group. (In fact, I have previously suggested, there should be a button "Do you need points?! Click this" and it hands out 340 points. This would immediately end the (wholly bizarre) problem of passersby who "want!! points!! on!! SO!!".) (Personally when I can be bothered I GIVE AWAY my points via bounties, since, you know, I'm not 7 years old playing minecraft for points...) | |
| Jan 3, 2023 at 0:34 | comment | added | danh | Oh, and would there be any reason to be less harsh with those who plagiarize from other humans? We see plenty of that on SO, often very easily detected. | |
| Jan 3, 2023 at 0:33 | comment | added | danh | I think you're describing a penalty that you'd like to see applied to those who have certainly plagiarized. The problem is, by making the stakes so high, you make the need for certainty proportionally high. What about a non-plagiarizing poor shmoe who answers with obsequious-seeming English and incorrect code? (I've probably done this, and whatever else you might think about me, you probably wouldn't want to see me banned). I think we'd want to have some milder punishment that escalates with instances of (suspected) robot-plagarism. | |
| Jan 1, 2023 at 16:04 | comment | added | Fattie | Of course, it's not my site. Others, perhaps a majority, may like the idea of cgpt output appearing on SO. And anyway, it's no democracy, it's just a business owned by a media group. | |
| Jan 1, 2023 at 15:50 | comment | added | Fattie | I'm afraid that's how I see it. If, incredibly, an "AI" system had arrived that can literally (as someone mentioned) read and understand documentation, track changes in (say) iOS, run experimental builds, synthesize existing clues, merge info from various fields, and come up with answers to actual (non-trivial) software questions - nobody would be a bigger fan than me. And if that happened the company that owns SO would have already dumped SO on to some other buyer. As it stands, the fluff on SO (non-experts rather pathetically trying to answer questions - who knows why) is just annoying. | |
| Jan 1, 2023 at 15:45 | comment | added | Fattie | (presumably they do this to get "points" on SO - I don't know or care why non-programmers add fluff to SO, but it's infuriating for those paid programmers who use SO, for "research questions", as part of their making a living to pay for eg. diapers.) Regarding your comments about "scaffolding" I have no idea what you may mean. Do you mean as a sort of spell checker? Or? Regarding trivial questions, google (or CGPT) will give you a "rough answer". You're then saying one uses grammarly or cgpt or something to help with the English? Regarding "research" questions, goog is useless. | |
| Jan 1, 2023 at 15:42 | comment | added | Fattie | ... CGPT is terrific for beginners answering the simplest questions. (Although utterly pointless, for beginners, because you just google for the identical info) But it's completely useless for (let's call it, for want of a better term) "research level questions" which is exactly what all programmers (other than just beginner hobbyists) use SO for. You may ask "why am I annoyed by people using CGPT on SO?" The reason is (as stated endlessly) a plague on SO is, beginner programmers with no real knowledge, "answering" questions, with no clue what they're talking about ... | |
| Jan 1, 2023 at 15:38 | comment | added | Fattie | Hi @TheMuffinMan . I have no interest in what anyone did or didn't "claim", I'm stating what I feel like stating, as does everyone in comments. Note that, I don't give AF about SO (I wish I owned shares in it!). over time, seemingly "forever" companies, even google, apple etc and certainly SO will fade away and new "forever" companies will take their place. Regarding your comments immediately above, I precisely said that CGPT is "merely google". There is, literally, nothing you a gain from it that you cannot gain from google. That means that (as has been said repeatedly), exactly like google, | |
| Jan 1, 2023 at 6:51 | comment | added | The Muffin Man | @Fattie Not a single person came in here and claimed that Chat GPT is the end all of AI. It's an extremely useful tool for saving people time. Claiming that it's useless is like saying that Google search is useless. For code, it gives you the ability to get some pretty robust scaffolding and the ability to continually modify it based on contextual understanding. We already have scaffolding tools, but this takes it to the next level. There's no reason this can't be used to "scaffold" an SO answer for someone. | |
| Jan 1, 2023 at 4:22 | history | edited | Fattie | CC BY-SA 4.0 |
added 5 characters in body
|
| Jan 1, 2023 at 4:21 | comment | added | Fattie | @MaxHodges - since chatgpt appeared, every single question I've needed to ask on SO, I've asked chatGPT first. in all cases it did not even produce coherent text. (Which is as one would expect, since - shock - there is no such thing as artificial intelligence yet, even vaguely, and if there was, it would be news on the scale of aliens arriving.) ChatGPT simply (effectively) googles, and then has a (excellent) engine which mimics the feel and grammar of an English corpus. Nobody would be more excited than me if there was an AI system that could answer SO questions | |
| Jan 1, 2023 at 2:16 | comment | added | maxhodges | there's already a system in place to downvote wrong answers. | |
| Dec 31, 2022 at 20:59 | comment | added | Fattie | Just BTW that answer (which it did nothing more than copy from a google search) is poor quality; there's no reason to place the braces like that. | |
| Dec 31, 2022 at 19:05 | comment | added | Fattie | Muff, it can only do trivial existing knowledge. | |
| Dec 31, 2022 at 9:15 | comment | added | The Muffin Man |
So I just tried it and got let radians = (degrees * .pi) / 180.0. This whole thread is so funny. SO days are numbered with this type of technology coming through.
|
|
| Dec 16, 2022 at 13:09 | comment | added | Fattie | @ChristophRackwitz - it's just infuriating. And the real solution is, all of us just need to tap delete more frequently. | |
| Dec 16, 2022 at 3:10 | comment | added | Christoph Rackwitz | I've witnessed users showing up, delivering low quality answers to the constant stream of duplicate low hanging fruit (instead of voting to close as duplicates), and promptly stopping as soon as they reach a round number of rep. | |
| Dec 11, 2022 at 6:15 | history | edited | starballMod | CC BY-SA 4.0 |
fix typo
|
| Dec 9, 2022 at 14:24 | history | edited | user438383 | CC BY-SA 4.0 |
added 216 characters in body
|
| Dec 9, 2022 at 12:49 | history | answered | Fattie | CC BY-SA 4.0 |