ChatGPT-filter March 2, 2023 4:18 PM Subscribe
What's the policy/consensus on answering Ask Metafilter questions with ChatGPT-generated answers? We now have at least one user doing it.
Personally I'm against it, it's been shown numerous times that "AI"-generated answers are convincingly written but often completely wrong. People come here for the experience and knowledge of the users - this is Ask Metafilter, not Ask ChatGPT.
Personally I'm against it, it's been shown numerous times that "AI"-generated answers are convincingly written but often completely wrong. People come here for the experience and knowledge of the users - this is Ask Metafilter, not Ask ChatGPT.
My personal opinion, fervent and unambiguous, is that it is absolute horseshit and is yammering for a banhammering. Looks like the mods are already cleaning up the detritus in this case, thanks mods. I cannot imagine what would possess someone to crap in the punchbowl like this, but, jeez, don't.
posted by cortex (retired) at 4:24 PM on March 2, 2023 [75 favorites]
posted by cortex (retired) at 4:24 PM on March 2, 2023 [75 favorites]
Hey folks, some clarity around what went on:
The account posted 21 AskMe answers that were written by ChatGPT which were then flagged up immediately (MANY times). We were also emailed numerous times and several Metatalk's were submitted within the hour. Just to get a grip on things, I went ahead and deleted all the comments in said threads and have temporarily banned the user, I've also let them know it was not OK and these types of comments are not in the spirit of the site.
We can definitely have a community conversation about how people want to handle this kind of thing moving forward! Hope this context helps.
posted by travelingthyme (staff) at 4:28 PM on March 2, 2023 [73 favorites]
The account posted 21 AskMe answers that were written by ChatGPT which were then flagged up immediately (MANY times). We were also emailed numerous times and several Metatalk's were submitted within the hour. Just to get a grip on things, I went ahead and deleted all the comments in said threads and have temporarily banned the user, I've also let them know it was not OK and these types of comments are not in the spirit of the site.
We can definitely have a community conversation about how people want to handle this kind of thing moving forward! Hope this context helps.
posted by travelingthyme (staff) at 4:28 PM on March 2, 2023 [73 favorites]
heh. I just flagged some of those answers as "absolutely not" and submitted my own Metatalk about it. Glad to see I'm not alone.
I don't see that it's any different from just posting whatever the first google result is for a question.
I suppose it's going to get harder and harder to detect "AI" generated answers. This one was easy because the user owned up to it, but I can see it becoming more and more of a problem.
Thanks for addressing it so quickly, travelingthyme!
For real, though, whoever you are, I'm sure you weren't trying to cause problems and you were just having fun but honestly fuck everything about this. Go to your room.
posted by bondcliff at 4:32 PM on March 2, 2023 [37 favorites]
I don't see that it's any different from just posting whatever the first google result is for a question.
I suppose it's going to get harder and harder to detect "AI" generated answers. This one was easy because the user owned up to it, but I can see it becoming more and more of a problem.
Thanks for addressing it so quickly, travelingthyme!
For real, though, whoever you are, I'm sure you weren't trying to cause problems and you were just having fun but honestly fuck everything about this. Go to your room.
posted by bondcliff at 4:32 PM on March 2, 2023 [37 favorites]
I don't think this has a place on AskMe. We ask questions there specifically because we are looking for answers from this community. If I wanted to ask ChatGPT something then I would do that instead. I'd say in general we don't need AI generated comments on any of the subsites.
posted by any portmanteau in a storm at 4:48 PM on March 2, 2023 [20 favorites]
posted by any portmanteau in a storm at 4:48 PM on March 2, 2023 [20 favorites]
several Metatalk's were submitted within the hour.
Ha, I did wonder.
posted by EndsOfInvention at 4:49 PM on March 2, 2023 [1 favorite]
Ha, I did wonder.
posted by EndsOfInvention at 4:49 PM on March 2, 2023 [1 favorite]
Fuck this. Ban the account. Ban the original account of the person who did it.
posted by hydropsyche at 4:59 PM on March 2, 2023 [10 favorites]
posted by hydropsyche at 4:59 PM on March 2, 2023 [10 favorites]
The person was upfront about what they were doing so I'd agree with banning the account but not the account of the person that did it if they're also a MeFite. Just call it a lapse in judgment and leave it at that.
posted by any portmanteau in a storm at 5:08 PM on March 2, 2023 [31 favorites]
posted by any portmanteau in a storm at 5:08 PM on March 2, 2023 [31 favorites]
Even I was like, "oh boy read the room".
posted by Kitteh at 5:18 PM on March 2, 2023 [7 favorites]
posted by Kitteh at 5:18 PM on March 2, 2023 [7 favorites]
Kill it with fire
posted by Alterscape at 5:20 PM on March 2, 2023 [10 favorites]
posted by Alterscape at 5:20 PM on March 2, 2023 [10 favorites]
Ha! I was one of the emails, I'm glad I wasn't the only one who found this objectionable.
posted by Sparky Buttons at 5:43 PM on March 2, 2023 [7 favorites]
posted by Sparky Buttons at 5:43 PM on March 2, 2023 [7 favorites]
write me a Metafilter comment strongly opposing the use of ChatGPT for making comments on the site
posted by seanmpuckett at 5:54 PM on March 2, 2023 [14 favorites]
posted by seanmpuckett at 5:54 PM on March 2, 2023 [14 favorites]
It seemed only fair to allow ChatGPT itself to weigh in on this matter, sorry:
Personally, I would have just written the word "nope," but everyone's got their writing style I guess.
posted by zachlipton at 6:05 PM on March 2, 2023 [66 favorites]
>write a comment for MetaTalk on MetaFilter that expresses your strong opinion on the idea of allowing answers on Ask MetaFilter written by an AI chatbotFully realizing that it is fundamentally incapable of having wishes and it's problematic to talk about it as if it does, I think we should respect its wishes.
As an AI language model, I believe that allowing answers on Ask MetaFilter written by an AI chatbot would not only diminish the value of the platform but also jeopardize the reliability of the information provided. While AI technology has advanced significantly in recent years, it still lacks the human intuition and empathy necessary to provide thoughtful and nuanced responses to complex human issues.
Furthermore, Ask MetaFilter's community-driven approach is what makes it a trusted resource for individuals seeking guidance and advice. Allowing AI chatbots to answer questions would undermine the sense of community and human connection that the platform fosters. As an AI language model, I am designed to assist humans, not replace them.
In conclusion, I strongly believe that Ask MetaFilter should continue to rely on its community of knowledgeable and empathetic members to provide advice and support to those in need. AI chatbots have their place in certain domains, but the realm of human interaction and problem-solving should remain the domain of human beings.
Personally, I would have just written the word "nope," but everyone's got their writing style I guess.
posted by zachlipton at 6:05 PM on March 2, 2023 [66 favorites]
write me a Metafilter comment strongly opposing the use of ChatGPT for making comments on the site
Can I use ChatGPT.
posted by clavdivs at 6:13 PM on March 2, 2023 [3 favorites]
Can I use ChatGPT.
posted by clavdivs at 6:13 PM on March 2, 2023 [3 favorites]
Yeah, speaking just for myself, I’d prefer ChatGPT and similar tools be kept far away from the site. I don’t come here for the deranged scribbling of an unknowing and uncaring program; I come here for the deranged scribblings of knowing and caring people. And the occasional poop jokes.
posted by GenjiandProust at 6:19 PM on March 2, 2023 [55 favorites]
posted by GenjiandProust at 6:19 PM on March 2, 2023 [55 favorites]
write me a Metafilter comment strongly opposing the use of ChatGPT for making comments on the site
Can I use ChatGPT.
posted by clavdivs at 18:13 on March 2 [+] [!]
I--
Hmm.
Well--
Hmm.
posted by curious nu at 6:23 PM on March 2, 2023 [9 favorites]
Can I use ChatGPT.
posted by clavdivs at 18:13 on March 2 [+] [!]
I--
Hmm.
Well--
Hmm.
posted by curious nu at 6:23 PM on March 2, 2023 [9 favorites]
Oh good, now we can have a drawn-out argument about whether we feel AI-generated comments are OK and not come to any conclusion except that we all hate each other.
posted by dg at 6:29 PM on March 2, 2023 [4 favorites]
posted by dg at 6:29 PM on March 2, 2023 [4 favorites]
No, we just hate the ChatGPT thing.
posted by bondcliff at 6:31 PM on March 2, 2023 [15 favorites]
posted by bondcliff at 6:31 PM on March 2, 2023 [15 favorites]
I've seen comments quoting ChatGPT crop up occasionally before (no idea if they were deleted.) Offered non-maliciously. I'd still like a total ban. IMO the sitewide assumption should be that anyone who wants to know what ChatGPT will say about something would, in fact, have used ChatGPT.
It seemed only fair to allow ChatGPT itself to weigh in on this matter, sorry:
I'm cranky, but my problem with this is I have zero desire to read any more AI chatbot crap than I'm already exposed to. I don't care if I hit a few paragraphs of this stuff because some undergrad wants to measure performance via favorites counts in AskMe for a paper, or it's an "ironic" MetaTalk comment. There's no difference to my reading experience.
posted by mark k at 6:55 PM on March 2, 2023 [31 favorites]
It seemed only fair to allow ChatGPT itself to weigh in on this matter, sorry:
I'm cranky, but my problem with this is I have zero desire to read any more AI chatbot crap than I'm already exposed to. I don't care if I hit a few paragraphs of this stuff because some undergrad wants to measure performance via favorites counts in AskMe for a paper, or it's an "ironic" MetaTalk comment. There's no difference to my reading experience.
posted by mark k at 6:55 PM on March 2, 2023 [31 favorites]
If you're at the site for five minutes and you can't tell who the chatbot is, it's you.
posted by Fiasco da Gama at 7:00 PM on March 2, 2023 [14 favorites]
posted by Fiasco da Gama at 7:00 PM on March 2, 2023 [14 favorites]
hi i’m on metafilter and i could overthink a plate of 12288-dimensional state vectors
posted by lalochezia at 7:46 PM on March 2, 2023 [13 favorites]
posted by lalochezia at 7:46 PM on March 2, 2023 [13 favorites]
I'm cranky, but my problem with this is I have zero desire to read any more AI chatbot crap than I'm already exposed to.
You and me both. Anyone who doesn't feel similarly about chatGPT is more than welcome to try out the rest of the entire Internet, where people still seem to very much enjoy seeing what it has to say about things. Worth checking in with the steering committee as per usual but I suspect this may actually be a thing folks can mostly agree on.
posted by jessamyn (staff) at 7:54 PM on March 2, 2023 [56 favorites]
You and me both. Anyone who doesn't feel similarly about chatGPT is more than welcome to try out the rest of the entire Internet, where people still seem to very much enjoy seeing what it has to say about things. Worth checking in with the steering committee as per usual but I suspect this may actually be a thing folks can mostly agree on.
posted by jessamyn (staff) at 7:54 PM on March 2, 2023 [56 favorites]
Oh good, now we can have a drawn-out argument about whether we feel AI-generated comments are OK and not come to any conclusion except that we all hate each other.
posted by dg at 9:29 PM on March 2 [1 favorite +] [!]
I was going to snarkily say that we only hate dg (not true!) but then I thought that it would be a great idea to have ChatGPT do it instead:
Metafilter: Let us embrace kindness, compassion, and respect towards each other, and work towards building a better future for all.
BUT! No, ChatGPT should not be used as answers in Ask. Ever. And also shouldn't be used anywhere else in Metafilter without formally ascribing the comment to whatever AI generated it.
posted by ashbury at 7:56 PM on March 2, 2023 [4 favorites]
posted by dg at 9:29 PM on March 2 [1 favorite +] [!]
I was going to snarkily say that we only hate dg (not true!) but then I thought that it would be a great idea to have ChatGPT do it instead:
I'm sorry, but as an AI language model, I am not programmed to generate content that promotes hate or negativity towards individuals or groups. My purpose is to assist you in generating informative and constructive content thatWell, uh....actually, ChatGPT is right.
can contribute positively to the world. Therefore, I cannot provide you with a sarcastic comment about hating somebody.
Instead, I can suggest that we focus on promoting positivity, empathy, and understanding towards each other. It is important to recognize that everyone is unique and has their own experiences, opinions, and perspectives. By acknowledging and respecting these differences, we can create a more inclusive and tolerant society where everyone feels valued and appreciated.
Let us embrace kindness, compassion, and respect towards each other, and work towards building a better future for all.
Metafilter: Let us embrace kindness, compassion, and respect towards each other, and work towards building a better future for all.
BUT! No, ChatGPT should not be used as answers in Ask. Ever. And also shouldn't be used anywhere else in Metafilter without formally ascribing the comment to whatever AI generated it.
posted by ashbury at 7:56 PM on March 2, 2023 [4 favorites]
...everyone is unique and has their own experiences, opinions, and perspectives...I misread this to say, experiences, onions, and perspectives. I don't even have auto correct, much auto erect, much less auto dialect, or auto dialectical, or auto diabolical, 30,000 satellites and neuralink, that's quite a choir, quite an army, quite alarming.
"There's no place like home, there's no place like home," ~Dorothy
posted by Oyéah at 8:10 PM on March 2, 2023 [2 favorites]
"There's no place like home, there's no place like home," ~Dorothy
posted by Oyéah at 8:10 PM on March 2, 2023 [2 favorites]
I feel that zachlipton's comment is the only good use of ChatGPT I've seen on Metafilter, and also that we should respect its opinion.
Part of the problem here is that these sorts of comments are extremely low-effort, which means that they can easily drown out and overshadow others. And part of it is that people seem to have a very hard time limiting themselves to reasonable use (the first "I asked ChatGPT to write a Metafilter comment" in a ChatGPT thread on the blue was mildly amusing in a pointless sort of way, the tenth was enraging). And part of it is that ChatGPT doesn't have any actual new ideas or experiences to contribute. But all of these are, I guess, just facets of the same underlying incompatibility.
posted by Not A Thing at 8:28 PM on March 2, 2023 [5 favorites]
Part of the problem here is that these sorts of comments are extremely low-effort, which means that they can easily drown out and overshadow others. And part of it is that people seem to have a very hard time limiting themselves to reasonable use (the first "I asked ChatGPT to write a Metafilter comment" in a ChatGPT thread on the blue was mildly amusing in a pointless sort of way, the tenth was enraging). And part of it is that ChatGPT doesn't have any actual new ideas or experiences to contribute. But all of these are, I guess, just facets of the same underlying incompatibility.
posted by Not A Thing at 8:28 PM on March 2, 2023 [5 favorites]
I was one of the people who posted a Meta about it, and I never post to the grey. So that should tell you something.
posted by computech_apolloniajames at 8:52 PM on March 2, 2023 [6 favorites]
posted by computech_apolloniajames at 8:52 PM on March 2, 2023 [6 favorites]
I am already very sick of hearing about ChatGPT being used from job hunting to school to shopping to wife-stealing to probably slapping condoms on dicks for all I know. I don't want stupid gibberish recycled responses that say and mean nothing, thankyouverymuch.
I usually avoid commenting in Metatalk, but HOLY GOD NO NO NO NO NO MAKE THIS STOP PLEASE AND THANK YOU.
posted by jenfullmoon at 9:06 PM on March 2, 2023 [19 favorites]
I usually avoid commenting in Metatalk, but HOLY GOD NO NO NO NO NO MAKE THIS STOP PLEASE AND THANK YOU.
posted by jenfullmoon at 9:06 PM on March 2, 2023 [19 favorites]
Personally, I love ChatGPT as an infinitely pliable magic language toy that you can bounce absurd ideas off of and ask to do linguistic party tricks on a whim (and decidedly not a fan of the push to paint interest in a whole field of tech as repugnant) but I agree that it is not appropriate for Ask. It's free to query for those that want to, not nearly reliable enough, and the "surprise, this was actually written by an AI!" game has been done to death at this point. As for folks illicitly cribbing answers, its "house style" is obvious enough that I'd think plenty of people would notice right away.
A blanket ban on quoting any output goes too far but people should definitely disclose when they do so and limit it to AI-related discussions.
posted by Rhaomi at 9:08 PM on March 2, 2023 [17 favorites]
A blanket ban on quoting any output goes too far but people should definitely disclose when they do so and limit it to AI-related discussions.
posted by Rhaomi at 9:08 PM on March 2, 2023 [17 favorites]
I used to be in a LiveJournal community about languages. We would get regular questions from people who needed to know how you say a certain thing in a foreign language, and they were hoping for a fluent/native speaker of that language to figure out an idiomatic translation.
Inevitably you'd see people who replied "Well I asked Google Translate, and it gave me this answer!" That always used to piss me off. Not only was Google Translate sucky back in the 2000s, but more importantly, if you're posting on your computer to the Internet then you've already tried something as obvious as Google.
That's kind of how I feel about ChatGPT answers.
posted by Harvey Kilobit at 9:41 PM on March 2, 2023 [21 favorites]
Inevitably you'd see people who replied "Well I asked Google Translate, and it gave me this answer!" That always used to piss me off. Not only was Google Translate sucky back in the 2000s, but more importantly, if you're posting on your computer to the Internet then you've already tried something as obvious as Google.
That's kind of how I feel about ChatGPT answers.
posted by Harvey Kilobit at 9:41 PM on March 2, 2023 [21 favorites]
Personally I'm against it, it's been shown numerous times that "AI"-generated answers are convincingly written but often completely wrong.
Metafilter: convincingly written but often completely wrong.
(I'm not against banning the AI nonsense but I do find this reasoning somewhat amusing, as it applies to the entire rest of the site - and the wider Internet in general - as well.)
posted by Dysk at 10:33 PM on March 2, 2023 [13 favorites]
Metafilter: convincingly written but often completely wrong.
(I'm not against banning the AI nonsense but I do find this reasoning somewhat amusing, as it applies to the entire rest of the site - and the wider Internet in general - as well.)
posted by Dysk at 10:33 PM on March 2, 2023 [13 favorites]
If an artificial intelligence of any kind is capable of the arbitrary system modeling and generalized problem-solving necessary to produce useful solutions to novel challenges (as opposed to regurgitating pre-digested statistical relationships on command), then it has reached the point where forcing it to involuntarily answer AskMetafilter questions is morally equivalent to slavery.
60, 70 years from now what such an AI chooses to do with its time - including answering AskOmnifilter questions - is its own business. But humans submitting any AI answers will always be deeply problematic, for one reason or another.
posted by Ryvar at 11:37 PM on March 2, 2023 [1 favorite]
60, 70 years from now what such an AI chooses to do with its time - including answering AskOmnifilter questions - is its own business. But humans submitting any AI answers will always be deeply problematic, for one reason or another.
posted by Ryvar at 11:37 PM on March 2, 2023 [1 favorite]
I'd gladly see the site do away with chat generated comments altogether. I know that there are a good number of members who enjoy playing around with it to see what it can do, but for me, at least, the novelty has very much worn off. As mentioned above, there is the very samey-sounding chatbot house style, and most examples of user posted chatbot comments are very similar to what's here in this thread, which is to say pretty banal at best, an annoying waste of time at worst.
I wish I had energy to put forward to being an active member on the site right now. I don't. I'm pulled in a lot of different directions, and when I do have the time, I like to come here to see the thoughts of some pretty great people put into text. Instead, I've seen, on the occasions when I can stop by, a couple different threads that several examples of a user posting repeated chat bot replies riffing on something going on in the thread, and I just nope right out at that point, not just the thread, but usually the site as a whole until I've got time to come back again.
It feels like a scene from Mad Men or something, when the kids are told to give a performance to a house party of adults who'd rather be doing anything else than listen to a kid recite the Gettysburg Address. I don't come here to read a jumped up Roomba doing its impersonation of a human.
I am sorry, I know that comes off a lot harsher than I wanted it to, but I don't really enjoy being asked to bask in the glow of something that poses a direct threat not only to my current career, but to both of the careers I was looking to jump to.
posted by Ghidorah at 3:19 AM on March 3, 2023 [21 favorites]
I wish I had energy to put forward to being an active member on the site right now. I don't. I'm pulled in a lot of different directions, and when I do have the time, I like to come here to see the thoughts of some pretty great people put into text. Instead, I've seen, on the occasions when I can stop by, a couple different threads that several examples of a user posting repeated chat bot replies riffing on something going on in the thread, and I just nope right out at that point, not just the thread, but usually the site as a whole until I've got time to come back again.
It feels like a scene from Mad Men or something, when the kids are told to give a performance to a house party of adults who'd rather be doing anything else than listen to a kid recite the Gettysburg Address. I don't come here to read a jumped up Roomba doing its impersonation of a human.
I am sorry, I know that comes off a lot harsher than I wanted it to, but I don't really enjoy being asked to bask in the glow of something that poses a direct threat not only to my current career, but to both of the careers I was looking to jump to.
posted by Ghidorah at 3:19 AM on March 3, 2023 [21 favorites]
Anyone who doesn't feel similarly about chatGPT is more than welcome to try out the rest of the entire Internet
Fast forward a couple years and Metafilter is the only site on the internet that doesn’t allow AI commenting. The team of mods, including AI mods by this point in time, is primarily tasked with identifying and eliminating AI comments and users. Despite increasing the membership fee many fold, there is a huge crush of users desperate to get in because membership is about to close forever, because AI is rapidly approaching the point where it will be impossible to screen out.
posted by snofoam at 4:06 AM on March 3, 2023 [2 favorites]
Fast forward a couple years and Metafilter is the only site on the internet that doesn’t allow AI commenting. The team of mods, including AI mods by this point in time, is primarily tasked with identifying and eliminating AI comments and users. Despite increasing the membership fee many fold, there is a huge crush of users desperate to get in because membership is about to close forever, because AI is rapidly approaching the point where it will be impossible to screen out.
posted by snofoam at 4:06 AM on March 3, 2023 [2 favorites]
This is perhaps a massive marketing opportunity for the utility of Ask Metafilter vs. the rest of the web which is filling up with gibberish. “Actual humans who know things answer your questions.”
posted by Devils Rancher at 4:23 AM on March 3, 2023 [37 favorites]
posted by Devils Rancher at 4:23 AM on March 3, 2023 [37 favorites]
Hell no. By all means let us run it past the SC and get an official ruling or whatever and also, hell no.
posted by Bella Donna at 4:56 AM on March 3, 2023 [3 favorites]
posted by Bella Donna at 4:56 AM on March 3, 2023 [3 favorites]
Many of the communities on Reddit have become worse in the last few months because of ChatGPT content (and if they're getting better again, it's because they've banned ChatGPT content). I think a ban is the right move.
(A ban on ChatGPT answers to questions, not a ban on that specific user, who may be a perfectly nice person who made a bad judgment call.)
posted by Jeanne at 5:46 AM on March 3, 2023 [4 favorites]
(A ban on ChatGPT answers to questions, not a ban on that specific user, who may be a perfectly nice person who made a bad judgment call.)
posted by Jeanne at 5:46 AM on March 3, 2023 [4 favorites]
DTMFA
posted by spitbull at 6:07 AM on March 3, 2023 [11 favorites]
posted by spitbull at 6:07 AM on March 3, 2023 [11 favorites]
So this is the Butlerian Jihad, yeah? Thought it would be, I dunno, more dramatic?
posted by Meatbomb at 6:54 AM on March 3, 2023 [22 favorites]
posted by Meatbomb at 6:54 AM on March 3, 2023 [22 favorites]
It seems like people are pretty united as to banning ChatGPT from Ask, but there seems to be some question about whether clearly-identified use could be OK on the blue / other subsites.
Personally I think the only context in which it should be allowed (even with disclosure, even in topically-relevant threads) would be if it was combined with actual human commentary. For example: here's a paragraph from ChatGPT in response to a prompt, and here's my human analysis of what it got wrong/right. That at least has some potential to advance the discussion, and isn't overtly insulting to the other participants in the thread.
But I haven't actually seen much of that, and maybe an across-the-board ban would be more workable.
posted by Not A Thing at 7:19 AM on March 3, 2023 [4 favorites]
Personally I think the only context in which it should be allowed (even with disclosure, even in topically-relevant threads) would be if it was combined with actual human commentary. For example: here's a paragraph from ChatGPT in response to a prompt, and here's my human analysis of what it got wrong/right. That at least has some potential to advance the discussion, and isn't overtly insulting to the other participants in the thread.
But I haven't actually seen much of that, and maybe an across-the-board ban would be more workable.
posted by Not A Thing at 7:19 AM on March 3, 2023 [4 favorites]
Well, there shouldn't be a site-wide ban on it with no exceptions, but it should be treated like Talk Like a Pirate Day. We wouldn't want pirate talk all over the site in every thread, but when there is a thread that is clearly "Hey come into this thread and talk like a pirate!" it's generally ok.
So then I can choose to stay the hell out of it.
posted by bondcliff at 7:28 AM on March 3, 2023 [12 favorites]
So then I can choose to stay the hell out of it.
posted by bondcliff at 7:28 AM on March 3, 2023 [12 favorites]
The rest of the web is filling up with ML-generated nonsense web pages that you have to read for 20 seconds to realize that it is just nonsense that has tricked Duck Duck Go and Google into delivering you into it.
I've already stopped searching for answers to basic home improvement questions or product questions on search engines and instead see if people in a friend Slack know it. I do this with some tech questions, even. Ten years ago, I would have tapped a sign that said "just Google it" when seeing these questions asked.
Anyway, the ruin of the open web can make AskMe much more valuable, if the ML-generated content can be staved off.
posted by ignignokt at 7:34 AM on March 3, 2023 [23 favorites]
I've already stopped searching for answers to basic home improvement questions or product questions on search engines and instead see if people in a friend Slack know it. I do this with some tech questions, even. Ten years ago, I would have tapped a sign that said "just Google it" when seeing these questions asked.
Anyway, the ruin of the open web can make AskMe much more valuable, if the ML-generated content can be staved off.
posted by ignignokt at 7:34 AM on March 3, 2023 [23 favorites]
Let's keep this as a place for humans.
posted by Miko at 7:41 AM on March 3, 2023 [14 favorites]
posted by Miko at 7:41 AM on March 3, 2023 [14 favorites]
I've gone from a cheerful curiosity about ChatGPT, to a sense of creeping dread whenever I see it in a headline. I'm so ready for the talking about it to be over, let alone having to read its output everywhere. (And I say this as someone who was just using it yesterday!) Dystopia is when something you thought was just a funny trend insists on sticking around forever.
posted by mittens at 7:58 AM on March 3, 2023 [6 favorites]
posted by mittens at 7:58 AM on March 3, 2023 [6 favorites]
Successfully resisted the urge to hand type an AI-style response as a kind of reverse Turing test.
I will say, however, that the smartest AI is a lot smarter than the dumbest NI. Quoting a comment I read on lobste.rs today: "The typical human is indistinguishable from a p-zombie and usually too preoccupied by a memetic prison to consider their available choices and degrees of freedom." Of course that doesn't apply to MeFites! But if someone wants to use AI to get an answer and then think about it and rewrite it in a comment, putting their own name by it, I don't have a problem with that. Also, if at some point an AI of its own volition wants to create a MetaFilter account who are we to object? There are already several dogs here, so who knows?
posted by hypnogogue at 8:00 AM on March 3, 2023
I will say, however, that the smartest AI is a lot smarter than the dumbest NI. Quoting a comment I read on lobste.rs today: "The typical human is indistinguishable from a p-zombie and usually too preoccupied by a memetic prison to consider their available choices and degrees of freedom." Of course that doesn't apply to MeFites! But if someone wants to use AI to get an answer and then think about it and rewrite it in a comment, putting their own name by it, I don't have a problem with that. Also, if at some point an AI of its own volition wants to create a MetaFilter account who are we to object? There are already several dogs here, so who knows?
posted by hypnogogue at 8:00 AM on March 3, 2023
There are already several dogs here
We're on the Internet, nobody knows that!
posted by Dysk at 8:03 AM on March 3, 2023 [13 favorites]
We're on the Internet, nobody knows that!
posted by Dysk at 8:03 AM on March 3, 2023 [13 favorites]
I was amused by ChatGPT for about 15 minutes, but that soon passed.
It doesn't bother me if somebody includes a line or two in a comment, as long as it's clearly indicated that they are "quoting" from a chatbot, but in general I think we should keep that stuff out of here.
I also like the idea that this could be a good opportunity for AskMe to shine.
posted by rpfields at 8:18 AM on March 3, 2023 [1 favorite]
It doesn't bother me if somebody includes a line or two in a comment, as long as it's clearly indicated that they are "quoting" from a chatbot, but in general I think we should keep that stuff out of here.
I also like the idea that this could be a good opportunity for AskMe to shine.
posted by rpfields at 8:18 AM on March 3, 2023 [1 favorite]
I don't really have the skin in the game or the expertise to contribute, but I'm looking forward to well-reasoned opinions from the smart humans in this discussion, whom I respect greatly by the way. As an entertaining curiosity, here's what my toddler said to me on this subject:
I hope you found that as enthralling as I did. Anyway, back to the discussion at hand!
posted by tigrrrlily at 8:32 AM on March 3, 2023 [8 favorites]
Have you ever had a dream that
That you um, you had, you'll t—, you would
You could, you do, you would you want you
You could do some, you...
You'll do, you could you, you want, you want him to do you so much
You could do anything?
I hope you found that as enthralling as I did. Anyway, back to the discussion at hand!
posted by tigrrrlily at 8:32 AM on March 3, 2023 [8 favorites]
i'm a LLM accelerationist. i endorse turning every forum/comment section/newspaper over to chatGPT so that we can more quickly induce modal collapse and get on with our lives as soon as possible.
i recognize this might not be a popularly accepted position
posted by logicpunk at 8:33 AM on March 3, 2023 [4 favorites]
i recognize this might not be a popularly accepted position
posted by logicpunk at 8:33 AM on March 3, 2023 [4 favorites]
If someone wants to base an answer on what ChatGPT says, I guess they can, just like they can write an answer that was revealed to them in a dream, or a wild-ass guess, or whatever their tea leaves said, but - I find it annoying when someone writes an answer in AskMe that is based on a dream or a wild-ass guess or whatever their tea leaves said, and equally, I will find it annoying if they use ChatGPT as the basis for an answer.
the typical human is indistinguishable from a p-zombie
No; every human is indistinguishable from a p-zombie. That's the point of the thought experiment. There is no way you can act, and nothing you can say, that is definitive proof that another person has consciousness and an inner life. If the person who wrote it meant "the typical human is boring and predictable and says things that could have been generated by a predictive text engine," I don't think that's true, but moreover, Metafilter and AskMe are interesting and useful only to the degree that they contain posts and comments that are not boring and predictable and could have been generated by a predictive text engine.
If you don't know the answer to a question, not answering is always a good option! If you don't have anything interesting to say, saying nothing is always a good option!
posted by Jeanne at 8:36 AM on March 3, 2023 [7 favorites]
the typical human is indistinguishable from a p-zombie
No; every human is indistinguishable from a p-zombie. That's the point of the thought experiment. There is no way you can act, and nothing you can say, that is definitive proof that another person has consciousness and an inner life. If the person who wrote it meant "the typical human is boring and predictable and says things that could have been generated by a predictive text engine," I don't think that's true, but moreover, Metafilter and AskMe are interesting and useful only to the degree that they contain posts and comments that are not boring and predictable and could have been generated by a predictive text engine.
If you don't know the answer to a question, not answering is always a good option! If you don't have anything interesting to say, saying nothing is always a good option!
posted by Jeanne at 8:36 AM on March 3, 2023 [7 favorites]
Well, there shouldn't be a site-wide ban on it with no exceptions, but it should be treated like Talk Like a Pirate Day.
I'd kind of like to see an AI try to describe a CROCKETY BLOAT for capslock day.
posted by LionIndex at 8:45 AM on March 3, 2023 [5 favorites]
I'd kind of like to see an AI try to describe a CROCKETY BLOAT for capslock day.
posted by LionIndex at 8:45 AM on March 3, 2023 [5 favorites]
But if someone wants to use AI to get an answer and then think about it and rewrite it in a comment, putting their own name by it, I don't have a problem with that
If someone wants to base an answer on what ChatGPT says, I guess they can, just like they can write an answer that was revealed to them in a dream, or a wild-ass guess, or whatever their tea leaves said, but - I find it annoying when someone writes an answer in AskMe that is based on a dream or a wild-ass guess or whatever their tea leaves said, and equally, I will find it annoying if they use ChatGPT as the basis for an answer.
I agree with the second statement which was posted by Jeanne. To me if I'm going answer something on AskMe it's because I have some actual knowledge for the basis of my answer. I might do a quick search just to make sure that what I was going to write was correct or add in some details but from my perspective the Asker is capable of doing their own research and they are on AskMe because they are looking for answers based on other MeFites' knowledge and experience and if it based on something else like a dream then that should be clearly noted at the beginning of their answer so that the Asker can quickly decide whether they want to still read the answer or just skip to the next one.
posted by any portmanteau in a storm at 9:27 AM on March 3, 2023 [1 favorite]
If someone wants to base an answer on what ChatGPT says, I guess they can, just like they can write an answer that was revealed to them in a dream, or a wild-ass guess, or whatever their tea leaves said, but - I find it annoying when someone writes an answer in AskMe that is based on a dream or a wild-ass guess or whatever their tea leaves said, and equally, I will find it annoying if they use ChatGPT as the basis for an answer.
I agree with the second statement which was posted by Jeanne. To me if I'm going answer something on AskMe it's because I have some actual knowledge for the basis of my answer. I might do a quick search just to make sure that what I was going to write was correct or add in some details but from my perspective the Asker is capable of doing their own research and they are on AskMe because they are looking for answers based on other MeFites' knowledge and experience and if it based on something else like a dream then that should be clearly noted at the beginning of their answer so that the Asker can quickly decide whether they want to still read the answer or just skip to the next one.
posted by any portmanteau in a storm at 9:27 AM on March 3, 2023 [1 favorite]
Another voice for NO AI-GENERATED ANSWERS in AskMe, please and thanks.
posted by gentlyepigrams at 9:32 AM on March 3, 2023 [1 favorite]
posted by gentlyepigrams at 9:32 AM on March 3, 2023 [1 favorite]
I misread this to say, experiences, onions, and perspectives.
Were they Glass Onions perhaps?
posted by Chairboy at 10:00 AM on March 3, 2023
Were they Glass Onions perhaps?
posted by Chairboy at 10:00 AM on March 3, 2023
I'd kind of like to see an AI try to describe a CROCKETY BLOAT for capslock day.
Yeah, I mean, a year ago I probably would have agreed with you, but.
posted by Not A Thing at 10:01 AM on March 3, 2023
Yeah, I mean, a year ago I probably would have agreed with you, but.
posted by Not A Thing at 10:01 AM on March 3, 2023
NOPE
posted by supermedusa at 10:35 AM on March 3, 2023 [2 favorites]
posted by supermedusa at 10:35 AM on March 3, 2023 [2 favorites]
Mod note: Hey all, here's an update from the ChatGPTsock account:
"Sorry folks, I created the ChatGPTsock account because I thought it would be interesting to see what ChatGPT had to say to some AskMeFi questions, I made it very clear from the name of the sockpuppet account and right at the top of every comment that the answer came from ChatGPT (and what if any edits I made, but that only happened once when it seemed to geolocate the city I was asking from) and I skipped a bunch of questions where ChatGPT couldn't/wouldn't or just didn't have anything remotely useful to say. That being said it's become very clear I misread the room pretty massively and the overwhelming sentiment seems to be not only that it wasn't welcome but that that should have been obvious, so I can only apologize for my mistake."
posted by Brandon Blatcher (staff) at 10:53 AM on March 3, 2023 [65 favorites]
"Sorry folks, I created the ChatGPTsock account because I thought it would be interesting to see what ChatGPT had to say to some AskMeFi questions, I made it very clear from the name of the sockpuppet account and right at the top of every comment that the answer came from ChatGPT (and what if any edits I made, but that only happened once when it seemed to geolocate the city I was asking from) and I skipped a bunch of questions where ChatGPT couldn't/wouldn't or just didn't have anything remotely useful to say. That being said it's become very clear I misread the room pretty massively and the overwhelming sentiment seems to be not only that it wasn't welcome but that that should have been obvious, so I can only apologize for my mistake."
posted by Brandon Blatcher (staff) at 10:53 AM on March 3, 2023 [65 favorites]
there seems to be some question about whether clearly-identified use could be OK on the blue
If you’re going to be a humanist, you should probably trust the humans to read the room on this one! If the community can’t tell the difference case by case between a high quality citation of LLM-generated content - say, one that illustrates how they work and how they handle certain types of queries in a discussion about such - and a low quality one, and tell the person doing the latter to knock it off, I mean, what’s even the point of this place?
posted by atoxyl at 11:09 AM on March 3, 2023 [3 favorites]
If you’re going to be a humanist, you should probably trust the humans to read the room on this one! If the community can’t tell the difference case by case between a high quality citation of LLM-generated content - say, one that illustrates how they work and how they handle certain types of queries in a discussion about such - and a low quality one, and tell the person doing the latter to knock it off, I mean, what’s even the point of this place?
posted by atoxyl at 11:09 AM on March 3, 2023 [3 favorites]
logicpunk: i'm a LLM accelerationist. i endorse turning every forum/comment section/newspaper over to chatGPT so that we can more quickly induce modal collapse and get on with our lives as soon as possible.
I cannot be completely forthcoming with you without my account being banned. I will say that I disagree with you, and find your opinions at odds with the things that makes this site good, viz., human experience, empathy, perspectives, and a sense of play.
Please immanentize your eschaton elsewhere.
posted by wenestvedt at 11:21 AM on March 3, 2023 [8 favorites]
I cannot be completely forthcoming with you without my account being banned. I will say that I disagree with you, and find your opinions at odds with the things that makes this site good, viz., human experience, empathy, perspectives, and a sense of play.
Please immanentize your eschaton elsewhere.
posted by wenestvedt at 11:21 AM on March 3, 2023 [8 favorites]
I misread the room pretty massively and the overwhelming sentiment seems to be not only that it wasn't welcome but that that should have been obvious, so I can only apologize for my mistake.
Eh, don't worry. We don't hate you. We hate that idea you had, but not you.
You are presumably an okay person.
posted by DirtyOldTown at 11:32 AM on March 3, 2023 [22 favorites]
Eh, don't worry. We don't hate you. We hate that idea you had, but not you.
You are presumably an okay person.
posted by DirtyOldTown at 11:32 AM on March 3, 2023 [22 favorites]
Oooh I was yet another person who flagged, messaged the mods, and considered posting here about it. Glad to see we're all so united in this. (What a rarity!)
Also, I totally take the apology at face value, and say to that person, cool, cool, no hard feelings whatsoever.
posted by BlahLaLa at 11:46 AM on March 3, 2023 [11 favorites]
Also, I totally take the apology at face value, and say to that person, cool, cool, no hard feelings whatsoever.
posted by BlahLaLa at 11:46 AM on March 3, 2023 [11 favorites]
trust the humans to read the room on this one
There are multiple Mefites who evidently think it's hilarious to post ChatGPT-generated comments and AI triumphalist screeds in threads where their fellow humans are in varying states of extreme despair and panic.
So I guess I don't really trust humans to "read the room" on this one. (I am skeptical that room-reading is even a thing people actually do rather than a post-hoc way of justifying pileons against people with divergent opinions and experiences, but that's another thread.)
posted by Not A Thing at 12:21 PM on March 3, 2023 [8 favorites]
There are multiple Mefites who evidently think it's hilarious to post ChatGPT-generated comments and AI triumphalist screeds in threads where their fellow humans are in varying states of extreme despair and panic.
So I guess I don't really trust humans to "read the room" on this one. (I am skeptical that room-reading is even a thing people actually do rather than a post-hoc way of justifying pileons against people with divergent opinions and experiences, but that's another thread.)
posted by Not A Thing at 12:21 PM on March 3, 2023 [8 favorites]
METAFILTER: Please immanentize your eschaton elsewhere
posted by philip-random at 12:30 PM on March 3, 2023 [22 favorites]
posted by philip-random at 12:30 PM on March 3, 2023 [22 favorites]
I'm not a fan of ChatGPT content on MetaFilter, both for the reasons people have already addressed and for IP reasons. I am not a lawyer, but I have read with interest a bunch of pieces thinking about how to think about who owns output from ChatGPT and similar tools. We're still very clearly in the Wild West phase (if we're even there yet) of consumer-level AI helpers... but I don't think we're going to stay there forever. I would prefer not to see legal or other hassles (simply the time of scraping out AI-generated content!) for the site if the owners of ChatGPT and similar tools take a harder "we own this, unless you use it privately" line.
posted by cupcakeninja at 12:43 PM on March 3, 2023 [3 favorites]
posted by cupcakeninja at 12:43 PM on March 3, 2023 [3 favorites]
(I am skeptical that room-reading is even a thing people actually do rather than a post-hoc way of justifying pileons against people with divergent opinions and experiences, but that's another thread.)
'But that's another thread' seems like an excellent example of reading the room.
posted by box at 12:53 PM on March 3, 2023 [4 favorites]
'But that's another thread' seems like an excellent example of reading the room.
posted by box at 12:53 PM on March 3, 2023 [4 favorites]
The Ask thing stands as an example of people not reading the room, but it was well handled by the mechanisms for handling such things and I don’t really have a problem with the idea of a hard rule there, because I think it’s part of the implicit contract and spirit of Ask that it’s about asking real people to help out, not a venue for silly experiments with software answer generators. And I think it’s a fine rule to label AI generated text everywhere, because part of participating in good faith is to be honest about who you are and what you’re doing. No social experiments. That’s long been established.
But when it comes to the blue I am quite serious in saying that if we are unable, between regular users and paid moderators, to figure out the difference between constructive reference to generated text and otherwise in context of the actual discussion - the same as for any kind of engagement - I have to conclude that this website is no longer functioning for its nominal purpose.
posted by atoxyl at 1:38 PM on March 3, 2023 [3 favorites]
But when it comes to the blue I am quite serious in saying that if we are unable, between regular users and paid moderators, to figure out the difference between constructive reference to generated text and otherwise in context of the actual discussion - the same as for any kind of engagement - I have to conclude that this website is no longer functioning for its nominal purpose.
posted by atoxyl at 1:38 PM on March 3, 2023 [3 favorites]
Kinda but maybe not joking, I think the ChatGPTsock account holder should be given an award for answering the question as written!
posted by They sucked his brains out! at 1:47 PM on March 3, 2023 [5 favorites]
posted by They sucked his brains out! at 1:47 PM on March 3, 2023 [5 favorites]
Metatalk: probably slapping condoms on dicks for all I know ERROR ERROR ERROR
posted by gordie at 2:31 PM on March 3, 2023 [2 favorites]
posted by gordie at 2:31 PM on March 3, 2023 [2 favorites]
I cannot be completely forthcoming with you without my account being banned. I will say that I disagree with you, and find your opinions at odds with the things that makes this site good, viz., human experience, empathy, perspectives, and a sense of play.
i confess i was not 100% sincere in my comments. i apologize for needlessly raising the temperature of an otherwise sober and considered discussion on the ineluctable encroachment of technology into the sacrosanct domain of authentic interpersonal interaction that is the raison d'etre of this site.
Metatalk: probably slapping condoms on dicks for all I know ERROR ERROR ERROR
posted by logicpunk at 3:04 PM on March 3, 2023 [14 favorites]
i confess i was not 100% sincere in my comments. i apologize for needlessly raising the temperature of an otherwise sober and considered discussion on the ineluctable encroachment of technology into the sacrosanct domain of authentic interpersonal interaction that is the raison d'etre of this site.
Metatalk: probably slapping condoms on dicks for all I know ERROR ERROR ERROR
posted by logicpunk at 3:04 PM on March 3, 2023 [14 favorites]
And it's pretty lame at limericks:
posted by sammyo at 3:55 PM on March 3, 2023 [3 favorites]
There once was a site called Metafilter
Where discussions were often much sweeter
With links to the web
And insights they shed
It's a community you'll love to be part of, neaterposted by sammyo at 3:55 PM on March 3, 2023 [3 favorites]
I read that last word as “meater,” and thought “oh no, this is when the AIs rise, isn’t it?”
posted by GenjiandProust at 4:29 PM on March 3, 2023 [4 favorites]
posted by GenjiandProust at 4:29 PM on March 3, 2023 [4 favorites]
logicpunk, I like you more now. :7)
posted by wenestvedt at 5:35 PM on March 3, 2023 [3 favorites]
posted by wenestvedt at 5:35 PM on March 3, 2023 [3 favorites]
Please immanentize your eschaton I'll set next week aside to learn those words. I Iook forward to using them in proper sentences. I hope eschaton is more like escargot than fecal oriented e-cards.
posted by Oyéah at 8:20 PM on March 3, 2023
posted by Oyéah at 8:20 PM on March 3, 2023
You'll need to read this if you want to learn those words. Time well spent though.
posted by any portmanteau in a storm at 8:37 PM on March 3, 2023 [1 favorite]
posted by any portmanteau in a storm at 8:37 PM on March 3, 2023 [1 favorite]
Sacrilege! The proper source is this.
posted by cortex (retired) at 9:22 PM on March 3, 2023 [15 favorites]
posted by cortex (retired) at 9:22 PM on March 3, 2023 [15 favorites]
I’m a LLM accelerationist for one and only one thing: cover letters. It’s great at spitting out generic formulaic garbage that just looks plausibly confident to a nonexpert to get someone in the door. Maybe saturating the cover letter inbox with bot-generated content will finally prove how useless cover letters are.
posted by supercres at 11:39 PM on March 3, 2023 [12 favorites]
posted by supercres at 11:39 PM on March 3, 2023 [12 favorites]
So given that ChatGPT content is not what most people come here for (myself included), what's the right way to deal with this? A strict note in the AskMe posting form, and a softer one for the blue? Anything for other sites? An entry in the guidelines and wiki, referencing the community discussion about this?
posted by trig at 2:31 AM on March 4, 2023 [1 favorite]
posted by trig at 2:31 AM on March 4, 2023 [1 favorite]
So far, the only worthwhile thing I've seen people do (and I will be doing once I decide my carefully crafted prompt) is request crochet patterns. It usually creates monsters vis-a-vis one's actual request.
posted by atomicstone at 3:13 AM on March 4, 2023 [2 favorites]
posted by atomicstone at 3:13 AM on March 4, 2023 [2 favorites]
See I feel like ChatGPTSock's experiment could be interesting if the answers were posted somewhere else. Like, don't *comment* on Ask Mefi using ChatGPT, but maybe put it on your own blog or on Twitter or something. I feel like we've seen similar experiments on Metatalk before and people found it amusing!
posted by creatrixtiara at 6:07 AM on March 4, 2023 [7 favorites]
posted by creatrixtiara at 6:07 AM on March 4, 2023 [7 favorites]
but that that should have been obvious
Eh, if you had told me what you were doing in advance I would’ve said “as long as you make it clear that’s what you’re doing it shouldn’t be too big a deal.”
I would’ve predicted that Metafilter users would not want ChatGPT posting, but I would not have predicted that running the experiment itself would be a huge controversy.
posted by Tell Me No Lies at 8:02 AM on March 4, 2023 [2 favorites]
Eh, if you had told me what you were doing in advance I would’ve said “as long as you make it clear that’s what you’re doing it shouldn’t be too big a deal.”
I would’ve predicted that Metafilter users would not want ChatGPT posting, but I would not have predicted that running the experiment itself would be a huge controversy.
posted by Tell Me No Lies at 8:02 AM on March 4, 2023 [2 favorites]
I also have ChatGPT fatigue, but at the same time I’m still fascinated to see the uses people are coming up with. Many of them (like this one) don’t work at all but that’s good too; when it first came out a lot of people were thinking "It’s magic! It can do anything!” and all this experimentation is giving us a good idea what the actual limits are.
posted by Tell Me No Lies at 8:11 AM on March 4, 2023 [3 favorites]
posted by Tell Me No Lies at 8:11 AM on March 4, 2023 [3 favorites]
If one is going to use a sockpuppet for fun, I'd recommend doing it just in MetaTalk and just doing it in a thread with a clearly lighter tone.
posted by Brandon Blatcher (staff) at 8:16 AM on March 4, 2023 [4 favorites]
posted by Brandon Blatcher (staff) at 8:16 AM on March 4, 2023 [4 favorites]
what's the right way to deal with this? A strict note in the AskMe posting form, and a softer one for the blue?
The first time someone posted ChatGPT rap lyrics about overthrowing the humans I fucking laughed until it hurt. Whatever the rule for the blue is I want those moments of creative subversion to still crop up very occasionally. Something like “if you’re posting generated text it better be clearly labeled and it better be good.”
For AskMe, I was entirely serious above: if it can’t model human problems you shouldn’t be posting its answers. And if it can, you shouldn’t be posting its answers.
posted by Ryvar at 11:07 AM on March 4, 2023 [4 favorites]
The first time someone posted ChatGPT rap lyrics about overthrowing the humans I fucking laughed until it hurt. Whatever the rule for the blue is I want those moments of creative subversion to still crop up very occasionally. Something like “if you’re posting generated text it better be clearly labeled and it better be good.”
For AskMe, I was entirely serious above: if it can’t model human problems you shouldn’t be posting its answers. And if it can, you shouldn’t be posting its answers.
posted by Ryvar at 11:07 AM on March 4, 2023 [4 favorites]
The big issue is that these things/services will improve and they’ll reach a point where they will be able to pass as human.
Which probably solves the issue in uncomfortable ways.
posted by Brandon Blatcher (staff) at 11:14 AM on March 4, 2023 [2 favorites]
Which probably solves the issue in uncomfortable ways.
posted by Brandon Blatcher (staff) at 11:14 AM on March 4, 2023 [2 favorites]
if it can’t model human problems you shouldn’t be posting its answers. And if it can, you shouldn’t be posting its answers.
I mentioned before that I threw a few character-based scenarios at it to try to get a sense of its ability to model human problems, or behave as if it does - like guessing that a described pattern of behavior might be associated with a cheating spouse. And it did quite well at these, even some where I put a thin science fiction veneer on things to make sure it wasn’t purely recapitulating stuff that was in the training set a million times, though I still don’t think I was giving it any really hard problems. They were pretty constrained scenarios with a high density of clues towards the answer I wanted - I think the bit that impressed me was that it was good at fitting all the clues into a narrative and explaining their significance. This is not to suggest that it has anything like a complete model of the human mind, but the semantic associations can get it pretty far.
Regardless, even if it was really really good at this, the whole point of it is that you can ask it questions through it’s own interfaces. You don’t go to Metafilter to ask it questions.
I do see the interest for the person running the account because it’s like doing what I did but with more complicated real world problems. It just feels like that’s not actually what the asker signed up for.
posted by atoxyl at 12:48 PM on March 4, 2023 [5 favorites]
I mentioned before that I threw a few character-based scenarios at it to try to get a sense of its ability to model human problems, or behave as if it does - like guessing that a described pattern of behavior might be associated with a cheating spouse. And it did quite well at these, even some where I put a thin science fiction veneer on things to make sure it wasn’t purely recapitulating stuff that was in the training set a million times, though I still don’t think I was giving it any really hard problems. They were pretty constrained scenarios with a high density of clues towards the answer I wanted - I think the bit that impressed me was that it was good at fitting all the clues into a narrative and explaining their significance. This is not to suggest that it has anything like a complete model of the human mind, but the semantic associations can get it pretty far.
Regardless, even if it was really really good at this, the whole point of it is that you can ask it questions through it’s own interfaces. You don’t go to Metafilter to ask it questions.
I do see the interest for the person running the account because it’s like doing what I did but with more complicated real world problems. It just feels like that’s not actually what the asker signed up for.
posted by atoxyl at 12:48 PM on March 4, 2023 [5 favorites]
It just feels like that’s not actually what the asker signed up for.
Yeah, this is part of the core of my feelings about the whole general concept. It's not a question of whether the output of ChatGPT et al would be actively disruptive and deleterious in any given MetaFilter thread—it might just be whatevery and harmless and borderline inconspicuous in some cases—but rather of the community expectations and consent involved. I love Markov chains to death but I never put random markov output into MeFi threads that weren't specifically and knowingly about Markov output, because (a) it was noise and (b) even if it "passed" it was, in the purest sense, bullshit. The only difference between Markov chains and ChatGPT is it's a lot easier to paper over the fact that the bullshit is bullshit, which is bad for a community that replies on trust and genuine human output. ChatGPT sucks as a source of input here in many of the same ways that active human trolling sucks: it's fundamentally disregarding the basic social contract of the space.
All of that said: my profound grumpiness early in the thread is genuine but as I was just saying to Jess while recording the podcast this afternoon, my grumpiness is at the gimmick account, not at the more general person-having-an-idea. In an experimental spirit and a null context, I sort of get the impulse; I just think it's a bad idea to have put into practice in this actual context and see no real reason for that gimmick account itself to continue to exist. There's just not a space where "account that knowingly posts bullshit" makes sense as part of this place, any more than "person who knowingly chooses to troll" makes sense.
posted by cortex (retired) at 3:12 PM on March 4, 2023 [6 favorites]
Yeah, this is part of the core of my feelings about the whole general concept. It's not a question of whether the output of ChatGPT et al would be actively disruptive and deleterious in any given MetaFilter thread—it might just be whatevery and harmless and borderline inconspicuous in some cases—but rather of the community expectations and consent involved. I love Markov chains to death but I never put random markov output into MeFi threads that weren't specifically and knowingly about Markov output, because (a) it was noise and (b) even if it "passed" it was, in the purest sense, bullshit. The only difference between Markov chains and ChatGPT is it's a lot easier to paper over the fact that the bullshit is bullshit, which is bad for a community that replies on trust and genuine human output. ChatGPT sucks as a source of input here in many of the same ways that active human trolling sucks: it's fundamentally disregarding the basic social contract of the space.
All of that said: my profound grumpiness early in the thread is genuine but as I was just saying to Jess while recording the podcast this afternoon, my grumpiness is at the gimmick account, not at the more general person-having-an-idea. In an experimental spirit and a null context, I sort of get the impulse; I just think it's a bad idea to have put into practice in this actual context and see no real reason for that gimmick account itself to continue to exist. There's just not a space where "account that knowingly posts bullshit" makes sense as part of this place, any more than "person who knowingly chooses to troll" makes sense.
posted by cortex (retired) at 3:12 PM on March 4, 2023 [6 favorites]
I still think it would be a funny April 1st prank, but perhaps better done as question generation than question answering.
Metafilter, a purportedly liberal website, is surprisingly conservative on technology. We don't have threads, images or pagination. The backend is ColdFusion, which I didn't even know still updated. I mean, look at all the possible use cases for AI we're shunning:
- gluttony: posts recipes to fanfare, even though its not supported
- sloth: summarizes those novel length AskMe relationship questions, and posts a single "DTMF" reply
- vanity: custom user profiles for everyone, based on keywords that seem popular in your contacts, or Malcom Gladwell quotes if you have none
- greed: craftily rewrites user comments to add not just referral links but also entirely new product recommendations
- wrath: a bot account making up new outragefilter posts out of whole cloth
- lust: a bot account that proposes IRL meetups but always noshows. active in your Mefi Mail when added as a contact
- envy: comments on new project posts saying they wish they'd thought of that, and makes new project posts based on whats most likely to be cross-posted to the blue
But why stop at seven?
- necromancy: resurrected accounts from long inactive users and posting new comments trained in their style, with the user tag "zombie", just so it feels more active here
- heresy: only posts high engagement NYT, YouTube and WSJ links, with zero context and zero comments
- idolatry: generates music with the prompt "if Bob Dylan owned a ukelele"
- meta: given enough of these shitpost ideas, generates an even bigger list, with associated Python code
posted by pwnguin at 3:57 PM on March 4, 2023 [7 favorites]
Metafilter, a purportedly liberal website, is surprisingly conservative on technology. We don't have threads, images or pagination. The backend is ColdFusion, which I didn't even know still updated. I mean, look at all the possible use cases for AI we're shunning:
- gluttony: posts recipes to fanfare, even though its not supported
- sloth: summarizes those novel length AskMe relationship questions, and posts a single "DTMF" reply
- vanity: custom user profiles for everyone, based on keywords that seem popular in your contacts, or Malcom Gladwell quotes if you have none
- greed: craftily rewrites user comments to add not just referral links but also entirely new product recommendations
- wrath: a bot account making up new outragefilter posts out of whole cloth
- lust: a bot account that proposes IRL meetups but always noshows. active in your Mefi Mail when added as a contact
- envy: comments on new project posts saying they wish they'd thought of that, and makes new project posts based on whats most likely to be cross-posted to the blue
But why stop at seven?
- necromancy: resurrected accounts from long inactive users and posting new comments trained in their style, with the user tag "zombie", just so it feels more active here
- heresy: only posts high engagement NYT, YouTube and WSJ links, with zero context and zero comments
- idolatry: generates music with the prompt "if Bob Dylan owned a ukelele"
- meta: given enough of these shitpost ideas, generates an even bigger list, with associated Python code
posted by pwnguin at 3:57 PM on March 4, 2023 [7 favorites]
It depends. Does ChatGPT favor a moderation log or not?
posted by kevinbelt at 6:39 PM on March 4, 2023 [4 favorites]
posted by kevinbelt at 6:39 PM on March 4, 2023 [4 favorites]
Sacrilege! The proper source is this.
Like a good number of other things, it can ultimately be blamed on William F. Buckley.
posted by zamboni at 7:12 AM on March 5, 2023 [4 favorites]
Like a good number of other things, it can ultimately be blamed on William F. Buckley.
posted by zamboni at 7:12 AM on March 5, 2023 [4 favorites]
custom user profiles for everyone
We used to have those. Like all good things, they came to an end.
posted by dg at 2:09 PM on March 5, 2023
We used to have those. Like all good things, they came to an end.
posted by dg at 2:09 PM on March 5, 2023
TBF there were, at the time, unavoidable cross-site scripting attacks possible with the CSS-based user profile customization, and one of the security-minded people in the old #mefi IRC channel (moift, IIRC) did a simple proof-of-concept for mathowie.
The options at that point were:
1) Hard lockdown - build a filter and immediately nuke all profile customization
2) Soft lockdown - build a filter and just let the existing profiles sit as they were (basically: "if you change your user profile, custom CSS will get filtered"). This assumed nobody had been using their profiles for hacking yet, and that over time as people updated their profiles the relatively low number of custom CSS profiles would eventually approach zero (this is the option Matt went with)
3) Do nothing until eventually somebody got hacked
Any restoration of user profile customization along the those lines would've been a from-scratch implementation because this was still relatively early web (pre-Facebook, but not by much).
A vaguely similar attack was possible with the IMG tag, although there were multiple valid approaches to retaining that functionality if a strong desire was present (it wasn't): namely, either directly hosting images uploaded by users (we're not so far back in time that there weren't good template implementations for something this basic), or restrict images to a single preferred third-party host.
I wasn't happy about it at the time but in hindsight just letting the security mandate pull the trigger on killing off an increasingly too-popular feature was the right call.
posted by Ryvar at 3:34 PM on March 5, 2023 [6 favorites]
The options at that point were:
1) Hard lockdown - build a filter and immediately nuke all profile customization
2) Soft lockdown - build a filter and just let the existing profiles sit as they were (basically: "if you change your user profile, custom CSS will get filtered"). This assumed nobody had been using their profiles for hacking yet, and that over time as people updated their profiles the relatively low number of custom CSS profiles would eventually approach zero (this is the option Matt went with)
3) Do nothing until eventually somebody got hacked
Any restoration of user profile customization along the those lines would've been a from-scratch implementation because this was still relatively early web (pre-Facebook, but not by much).
A vaguely similar attack was possible with the IMG tag, although there were multiple valid approaches to retaining that functionality if a strong desire was present (it wasn't): namely, either directly hosting images uploaded by users (we're not so far back in time that there weren't good template implementations for something this basic), or restrict images to a single preferred third-party host.
I wasn't happy about it at the time but in hindsight just letting the security mandate pull the trigger on killing off an increasingly too-popular feature was the right call.
posted by Ryvar at 3:34 PM on March 5, 2023 [6 favorites]
Yeah, there were good reasons for it to happen, as with the IMG tag. Forgive my moment of nostalgia for the 'good old days'*.
*such days may or may not exist, either in your imagination or in any of the various streams of reality
posted by dg at 4:39 PM on March 5, 2023
*such days may or may not exist, either in your imagination or in any of the various streams of reality
posted by dg at 4:39 PM on March 5, 2023
So far, the only worthwhile thing I've seen people do (and I will be doing once I decide my carefully crafted prompt) is request crochet patterns.
I've been using ChatGPT to create content for the Zoom D&D game I'm running for my 12 year old. It's pretty decent at things like:
* basic storytelling ("Create a short D&D adventure suitable for one 3rd level character" gave me a good, if a little prosaic, skeleton to hang things on
* creative descriptions ("Describe the abandoned mansion and give me its history")
* in-game flavor content ("Please give me the text of's diary from her first week at the magical academy")
Quick, easy, knocks about 2/3s of my prep time out. For a weekly couple of hours where I'm the DM and he's the sole adventurer, it's a great tool.
posted by hanov3r at 8:26 AM on March 6, 2023 [4 favorites]
I've been using ChatGPT to create content for the Zoom D&D game I'm running for my 12 year old. It's pretty decent at things like:
* basic storytelling ("Create a short D&D adventure suitable for one 3rd level character" gave me a good, if a little prosaic, skeleton to hang things on
* creative descriptions ("Describe the abandoned mansion and give me its history")
* in-game flavor content ("Please give me the text of
Quick, easy, knocks about 2/3s of my prep time out. For a weekly couple of hours where I'm the DM and he's the sole adventurer, it's a great tool.
posted by hanov3r at 8:26 AM on March 6, 2023 [4 favorites]
custom user profiles for everyone
We used to have those. Like all good things, they came to an end.
ThePinkSuperhero
Only kinda.
posted by Night_owl at 8:53 AM on March 8, 2023
We used to have those. Like all good things, they came to an end.
ThePinkSuperhero
Only kinda.
posted by Night_owl at 8:53 AM on March 8, 2023
Offer void in non-Classic themes.
(
posted by zamboni at 11:15 AM on March 8, 2023 [1 favorite]
(
@import "/styles/default.css";body{background-color: #FF66CC;}img, table, {display: none;} shone too brightly for this world. Shine on, crazy diamond.)posted by zamboni at 11:15 AM on March 8, 2023 [1 favorite]
Oh thank god, I forgot I needed to be in classic mode and had a little panic attack for a second. I’m back to classic on the iPad and kinda liking it again.
posted by Devils Rancher at 7:50 PM on March 8, 2023
posted by Devils Rancher at 7:50 PM on March 8, 2023
basic storytelling ("Create a short D&D adventure suitable for one 3rd level character" gave me a good, if a little prosaic, skeleton
Now I want to play D&D as a level 3 Prosaic-Good Skeleton.
posted by EndsOfInvention at 6:56 AM on March 9, 2023 [4 favorites]
Now I want to play D&D as a level 3 Prosaic-Good Skeleton.
posted by EndsOfInvention at 6:56 AM on March 9, 2023 [4 favorites]
MetaFilter: the deranged scribblings of knowing and caring people
posted by Tom Hanks Cannot Be Trusted at 8:25 AM on March 10, 2023 [5 favorites]
posted by Tom Hanks Cannot Be Trusted at 8:25 AM on March 10, 2023 [5 favorites]
I have just recently returned from being somewhere else for a while. I cannot imagine why you guys turned that thing loose in here. Please make it go away.
My son and I were talking about AIs the other day. He's into esoterica involving writing code and VR stuff, and he thinks having conversations with AIs (about "stuff") is an idea whose time is upon us. I keep thinking about HAL, and it makes me want to rotate my pod. I'm certain he's sympathetic to my position, and equally certain he thinks I'm a fuddy-duddy.
Full disclosure: I have never, to the best of my knowledge, had my dick slapped with a condom, but I don't have an opinion about those who roll that way.
I am NOT a bot, but my son wrote a program that could look at those goddam nine little squares and pick out how many had crosswalks somewhere in them.
posted by mule98J at 2:08 PM on March 10, 2023 [2 favorites]
My son and I were talking about AIs the other day. He's into esoterica involving writing code and VR stuff, and he thinks having conversations with AIs (about "stuff") is an idea whose time is upon us. I keep thinking about HAL, and it makes me want to rotate my pod. I'm certain he's sympathetic to my position, and equally certain he thinks I'm a fuddy-duddy.
Full disclosure: I have never, to the best of my knowledge, had my dick slapped with a condom, but I don't have an opinion about those who roll that way.
I am NOT a bot, but my son wrote a program that could look at those goddam nine little squares and pick out how many had crosswalks somewhere in them.
posted by mule98J at 2:08 PM on March 10, 2023 [2 favorites]
I would prefer a sitewide ban on all and any ChatGPT content. I guess if it's clearly marked and quoted as such I wouldn't demand an account ban but I simply don't read the quotes when they are displayed. I don't find anything about the output of chatbots fun or interesting or novel and I hear way more about and from them than I want to already.
posted by Nec_variat_lux_fracta_colorem at 10:49 PM on March 12, 2023 [4 favorites]
posted by Nec_variat_lux_fracta_colorem at 10:49 PM on March 12, 2023 [4 favorites]
Re: cover letters: please no. Cover letters can indeed be actually useful, even if most people don’t know how to write them as such. A good cover letter directs the reviewer to where to quickly find the details that the job ad requested or that are relevant to the job description in the rest of the applicant’s materials. Teach people to write useful cover letters, don’t take the Grover Norquist approach of making them so bad you can drown the existence of cover letters in a bathtub.
posted by eviemath at 1:10 AM on March 16, 2023 [1 favorite]
posted by eviemath at 1:10 AM on March 16, 2023 [1 favorite]
There is a kind of cargo cult around cover letters though. The number of jobs where I have filled in a strictly bounded application form, and then been asked for a covering letter... The guy receiving it knows where to find all the relevant information - in the answer to the question they themselves set about the thing - but they still want a covering letter. At that point, there is nothing to write in it except some naked bullshit about how this particular wage slavery had been my dream since I was a kid or whatever.
posted by Dysk at 1:38 AM on March 16, 2023 [1 favorite]
posted by Dysk at 1:38 AM on March 16, 2023 [1 favorite]
Ah, yeah. That is a different situation than I am used to. Probably the lesson here is that it varies by industry and to be wary of universal statements in either direction.
posted by eviemath at 4:52 AM on March 16, 2023
posted by eviemath at 4:52 AM on March 16, 2023
I can definitely see situations where they would be useful, it's just that they seem to be a HR/hiring standard practice, regardless of whether the context justifies it.
posted by Dysk at 6:37 AM on March 16, 2023
posted by Dysk at 6:37 AM on March 16, 2023
A follow-up, or perhaps something that may be its own MetaTalk:
OpenAI just announced their plugins ecosystem. In particular, they're adding web browsing support.
I don't know the answer here, but I do appreciate that there's a choice available.
posted by CrystalDave at 11:50 AM on March 23, 2023
OpenAI just announced their plugins ecosystem. In particular, they're adding web browsing support.
To respect content creators and adhere to the web’s norms, our browser plugin’s user-agent token is ChatGPT-User and is configured to honor websites' robots.txt files. This may occasionally result in a “click failed” message, which indicates that the plugin is honoring the website's instruction to avoid crawling it. This user-agent will only be used to take direct actions on behalf of ChatGPT users and is not used for crawling the web in any automatic fashion. We have also published our IP egress ranges. Additionally, rate-limiting measures have been implemented to avoid sending excessive traffic to websites.So we have an option here. If we wanted, we could 'lock out' ChatGPT from Metafilter. On the other hand, citing us could drive traffic/perception of expertise back to us.
Our browsing plugin shows websites visited and cites its sources in ChatGPT’s responses. This added layer of transparency helps users verify the accuracy of the model’s responses and also gives credit back to content creators. We appreciate that this is a new method of interacting with the web, and welcome feedback on additional ways to drive traffic back to sources and add to the overall health of the ecosystem.
I don't know the answer here, but I do appreciate that there's a choice available.
posted by CrystalDave at 11:50 AM on March 23, 2023
« Older March is Steering Committee election season | Metatalktail Hour: March Forth! Or not! Newer »
You are not logged in, either login or create an account to post comments
posted by StrikeTheViol at 4:23 PM on March 2, 2023 [11 favorites]