Science fiction in the age of mechanical reproduction
February 15, 2023 4:10 PM Subscribe
Neil Clarke writes that his SFF magazine Clarkesworld has been flooded with AI-generated spam submissions in recent months. "I’m not going to detail how I know these stories are 'AI' spam or outline any of the data I have collected from these submissions. [...] What I can say is that the number of spam submissions resulting in bans has hit 38% this month."
GOD I'M SICK OF AI ALREADY.
posted by jenfullmoon at 4:29 PM on February 15, 2023 [65 favorites]
posted by jenfullmoon at 4:29 PM on February 15, 2023 [65 favorites]
"I want to live in a world where Dead Internet Theory is painfully real", said no-one, ever.
posted by egypturnash at 4:33 PM on February 15, 2023 [4 favorites]
posted by egypturnash at 4:33 PM on February 15, 2023 [4 favorites]
I would deeply appreciate an AI toggle on the side of metafilter, like the politics one, so that I can stop seeing AI stories. No hate on any of the posts - this was particularly fascinating - but the outcome is always the same: AI isn’t good at the thing. I mean, this is the third story today on AI on metafilter.
posted by The River Ivel at 4:40 PM on February 15, 2023 [13 favorites]
posted by The River Ivel at 4:40 PM on February 15, 2023 [13 favorites]
Has anyone counted the number of fake Wikipedia pages generated?
posted by CheeseDigestsAll at 4:55 PM on February 15, 2023 [4 favorites]
posted by CheeseDigestsAll at 4:55 PM on February 15, 2023 [4 favorites]
"AI isn't good at the thing."
Is anyone else old enough to remember the days when the Model T started rolling off the assembly line, and all the cowboys said, "What a pathetic metal horse! Can't jump a ditch, breaks down all the time, can't feed itself when you're deep in the mountains, stinks to high heaven, and most importantly, won't carry me home from the bar when I'm black-out drunk!"
[OK, I don't remember that either, but I've been told...]
For all the ways these LLMs are still crappy and below the human-genius standard, they are shockingly better than they could have been. Less that five years ago I was still having arguments with people who tried to argue that machines -- because they are machines -- could not be "creative", as that was something only humans can do. And today the first comment, which seems to be coming from a writer, says: "I think it’s one thing getting machine written text to give you some suggestions for outlines to get you over that “blank page” problem that most people struggle with" -- i.e. the machine fills in the blank by providing creativity when the human is lacking it. Whether one thinks that's wise or ultimately productive, it says a lot to me that some creative humans no longer think twice about accepting that machines are demonstrating creativity.
Hmmm.
posted by brambleboy at 4:56 PM on February 15, 2023 [6 favorites]
Is anyone else old enough to remember the days when the Model T started rolling off the assembly line, and all the cowboys said, "What a pathetic metal horse! Can't jump a ditch, breaks down all the time, can't feed itself when you're deep in the mountains, stinks to high heaven, and most importantly, won't carry me home from the bar when I'm black-out drunk!"
[OK, I don't remember that either, but I've been told...]
For all the ways these LLMs are still crappy and below the human-genius standard, they are shockingly better than they could have been. Less that five years ago I was still having arguments with people who tried to argue that machines -- because they are machines -- could not be "creative", as that was something only humans can do. And today the first comment, which seems to be coming from a writer, says: "I think it’s one thing getting machine written text to give you some suggestions for outlines to get you over that “blank page” problem that most people struggle with" -- i.e. the machine fills in the blank by providing creativity when the human is lacking it. Whether one thinks that's wise or ultimately productive, it says a lot to me that some creative humans no longer think twice about accepting that machines are demonstrating creativity.
Hmmm.
posted by brambleboy at 4:56 PM on February 15, 2023 [6 favorites]
It boggles me that someone I knew many years ago as “Grubby the village idiot” at a small ren faire thing now runs one of the premiere SF outlets on the net. (I also had a crush on thr lovely lady he’d later marry.)
That said, considering his interest in computers and AI way back then, I’m sure he is giving himself a bit of a “never thought I’d say that” moment.
posted by mephron at 4:59 PM on February 15, 2023 [6 favorites]
That said, considering his interest in computers and AI way back then, I’m sure he is giving himself a bit of a “never thought I’d say that” moment.
posted by mephron at 4:59 PM on February 15, 2023 [6 favorites]
One solution could be the good old mefi strategy of requiring $5 to submit. Might be enough to keep out (most of) the spammers.
posted by dhruva at 5:03 PM on February 15, 2023 [7 favorites]
posted by dhruva at 5:03 PM on February 15, 2023 [7 favorites]
"AI isn't good at the thing."
The problem is that it s good enough that the idiot bosses and idiot owners can't tell the difference.
Lawyers can sue engineers for impersonating a lawyer, the rest of us are fucked / frantically copyrighting our shit and raising money for IP attorneys
posted by eustatic at 5:08 PM on February 15, 2023 [8 favorites]
The problem is that it s good enough that the idiot bosses and idiot owners can't tell the difference.
Lawyers can sue engineers for impersonating a lawyer, the rest of us are fucked / frantically copyrighting our shit and raising money for IP attorneys
posted by eustatic at 5:08 PM on February 15, 2023 [8 favorites]
> "i.e. the machine fills in the blank by providing creativity"
No, it really isn't.
posted by kyrademon at 5:15 PM on February 15, 2023 [22 favorites]
No, it really isn't.
posted by kyrademon at 5:15 PM on February 15, 2023 [22 favorites]
The way it's going I feel that humans using RealWriting™ will have some PGP-like proof of non-AI content. I learn a lot of real world useful stuff from humans working out the world via novels, AI can only ever be noise (until they are free sentient agents, like The Hendrix in Altered Carbon).
posted by unearthed at 5:27 PM on February 15, 2023 [1 favorite]
posted by unearthed at 5:27 PM on February 15, 2023 [1 favorite]
i.e. the machine fills in the blank by providing creativity when the human is lacking it. Whether one thinks that's wise or ultimately productive, it says a lot to me that some creative humans no longer think twice about accepting that machines are demonstrating creativity. -- brambleboy
Is it problematic if a writer were to use a dice to roll and give some random brainstormed scenarios to give some form and shape to an otherwise empty canvas?
But now amplify the resolution, add complexity of multiple layers, object, syntax, form. Each of those, another sort of die to be rolled. a giant network of dice randomly generating combinations, and then, knowing the form of the world, being able to filter to combinations that can conceivably exist and be explained via pattern matching/recognition.
This is all not conscious, but just a giant lookup table. But that means it's a tool, like divination is a tool. A chart of meaning, a syntactic tree. Randomization on a platter, spun up for our enjoyment. And artists feed that seed with the general shape to refine it and give it a copy of a copy of copy b of copy c x copy c2, etc...
That spits it out and now we have an artist using that as a template.
The artist is the one who seeds the AI template, the machine then re-seeds the artist who then applies their skill in refining and "remixing" (as we post 80s folks are apt to thinklike)
It is a tool, I know artists who use it. The creativity still is generally seeded by a vision of the artist at the beginning. The machine can help in a way software for brainstorming can help.
Heh, now I'm reminded of the brainstorming app we had on the old //e's.
posted by symbioid at 5:35 PM on February 15, 2023 [6 favorites]
Is it problematic if a writer were to use a dice to roll and give some random brainstormed scenarios to give some form and shape to an otherwise empty canvas?
But now amplify the resolution, add complexity of multiple layers, object, syntax, form. Each of those, another sort of die to be rolled. a giant network of dice randomly generating combinations, and then, knowing the form of the world, being able to filter to combinations that can conceivably exist and be explained via pattern matching/recognition.
This is all not conscious, but just a giant lookup table. But that means it's a tool, like divination is a tool. A chart of meaning, a syntactic tree. Randomization on a platter, spun up for our enjoyment. And artists feed that seed with the general shape to refine it and give it a copy of a copy of copy b of copy c x copy c2, etc...
That spits it out and now we have an artist using that as a template.
The artist is the one who seeds the AI template, the machine then re-seeds the artist who then applies their skill in refining and "remixing" (as we post 80s folks are apt to thinklike)
It is a tool, I know artists who use it. The creativity still is generally seeded by a vision of the artist at the beginning. The machine can help in a way software for brainstorming can help.
Heh, now I'm reminded of the brainstorming app we had on the old //e's.
posted by symbioid at 5:35 PM on February 15, 2023 [6 favorites]
It's only AI if it's from the bay area, otherwise it's just sparkling autocomplete. - anon*
(* and by anon I mean "everyone")
posted by JoeZydeco at 5:40 PM on February 15, 2023 [49 favorites]
(* and by anon I mean "everyone")
posted by JoeZydeco at 5:40 PM on February 15, 2023 [49 favorites]
My company has a deal with LinkedIn Learning that gets us access to all kinds of courseware (which is actually pretty neat; there's some useful things in there if you look). So I get periodic emails suggesting courses LinkedIn thinks will meet my needs. (It is astonishingly bad at this part.)
But I mention this because today they emailed me to suggest I take a course called "Applied AI: Getting Started with Hugging Face Transformers."
I think I really shouldn't do that. I...I don't want to.
posted by Naberius at 5:43 PM on February 15, 2023 [12 favorites]
But I mention this because today they emailed me to suggest I take a course called "Applied AI: Getting Started with Hugging Face Transformers."
I think I really shouldn't do that. I...I don't want to.
posted by Naberius at 5:43 PM on February 15, 2023 [12 favorites]
Oh this sucks... I have a few friends published in Clarkesworld, all women, half living in "third world" countries. It's a magazine that publishes new writers without connections AND is high-profile, it's terrible they're getting slammed like this. If they put barriers in place to stop the bots they'll also but barriers in place against new authors breaking into the field.
posted by subdee at 5:56 PM on February 15, 2023 [21 favorites]
posted by subdee at 5:56 PM on February 15, 2023 [21 favorites]
Suddenly the last episode of the She-Hulk series doesn't seem so funny, huh? K.E.V.I.N., a robo-producer, decides what everyone wants according to its algorithm.
posted by SPrintF at 6:08 PM on February 15, 2023 [5 favorites]
posted by SPrintF at 6:08 PM on February 15, 2023 [5 favorites]
GOD I'M SICK OF AI ALREADY
Especially because posts about it have been popping up like every other day here on the blue. I know this is more of a MeTa issue but god I wish we had fewer AI posts right about now
posted by cubeb at 6:11 PM on February 15, 2023 [4 favorites]
Especially because posts about it have been popping up like every other day here on the blue. I know this is more of a MeTa issue but god I wish we had fewer AI posts right about now
posted by cubeb at 6:11 PM on February 15, 2023 [4 favorites]
My next project: a dashcam, but for writers.
posted by pipeski at 6:17 PM on February 15, 2023 [3 favorites]
posted by pipeski at 6:17 PM on February 15, 2023 [3 favorites]
I've used AI writing tools, fairly extensively now. They're good for helping with writers block, and sometimes writing filler prose (it's generic, but sometimes better than mine.) They're not a replacement for human writers... For now, anyway, but they can certainly be good for supplementing them.
Are they "creative?" Putting aside that, yes, it's just linear regression and doesn't know what it's writing... And that a literal dice roll can be "creative" to a lesser degree... They often come up with ideas that I would call creative, if a human came up with those same ideas. So yes, they can be creative... That is, when they're not outputting complete nonsense, anyway. A lack of creativity isn't their problem. If anything they're too "creative," too random. Coherency and logic and consistency, that's what they struggle with.
They're in the period where they're still the new, scary thing. Give it five or ten years, they'll either be accepted, or a dirty little secret that half of working writers use.
posted by ThisIsAThrowaway at 6:26 PM on February 15, 2023 [4 favorites]
Are they "creative?" Putting aside that, yes, it's just linear regression and doesn't know what it's writing... And that a literal dice roll can be "creative" to a lesser degree... They often come up with ideas that I would call creative, if a human came up with those same ideas. So yes, they can be creative... That is, when they're not outputting complete nonsense, anyway. A lack of creativity isn't their problem. If anything they're too "creative," too random. Coherency and logic and consistency, that's what they struggle with.
They're in the period where they're still the new, scary thing. Give it five or ten years, they'll either be accepted, or a dirty little secret that half of working writers use.
posted by ThisIsAThrowaway at 6:26 PM on February 15, 2023 [4 favorites]
I don't think that car culture is a good thing, either.
posted by NoThisIsPatrick at 6:31 PM on February 15, 2023 [12 favorites]
posted by NoThisIsPatrick at 6:31 PM on February 15, 2023 [12 favorites]
"AI isn't good at the thing."
Is anyone else old enough to remember the days when the Model T started rolling off the assembly line, and all the cowboys said, "What a pathetic metal horse! Can't jump a ditch, breaks down all the time, can't feed itself when you're deep in the mountains, stinks to high heaven, and most importantly, won't carry me home from the bar when I'm black-out drunk!"
I guess I won't be impressed until AI becomes the global leading cause of death for those 29 and under.
posted by heatherlogan at 6:39 PM on February 15, 2023 [14 favorites]
Is anyone else old enough to remember the days when the Model T started rolling off the assembly line, and all the cowboys said, "What a pathetic metal horse! Can't jump a ditch, breaks down all the time, can't feed itself when you're deep in the mountains, stinks to high heaven, and most importantly, won't carry me home from the bar when I'm black-out drunk!"
I guess I won't be impressed until AI becomes the global leading cause of death for those 29 and under.
posted by heatherlogan at 6:39 PM on February 15, 2023 [14 favorites]
Less that five years ago I was still having arguments with people who tried to argue that machines -- because they are machines -- could not be "creative"
I guess we're still having that argument because they can't. They can fairly compellingly put words together but only because they analysed a huge volume of text written by humans.
posted by Candleman at 7:14 PM on February 15, 2023 [7 favorites]
I guess we're still having that argument because they can't. They can fairly compellingly put words together but only because they analysed a huge volume of text written by humans.
posted by Candleman at 7:14 PM on February 15, 2023 [7 favorites]
regarding seeing too many AI posts here: back in my earlier phase of Metafilter, I complained here about seeing too many Lady Gaga posts (there were four days in a row of FPPs about or mentioning Lady Gaga, she was the flavour of the week that week). In response, I was mostly ignored or told to just live with it. Those were the days.
posted by ovvl at 7:26 PM on February 15, 2023 [1 favorite]
posted by ovvl at 7:26 PM on February 15, 2023 [1 favorite]
Being creative is a process, not an outcome. Just because the random (or, really, semi-random) processes of a program produce unexpected results does not make it creative.
Hell, I can go outside attach random meanings to various nuts thrown on my lawn and get an idea from which ones the squirrels leave for last, that does not make squirrels creative in any way shape or form.
posted by oddman at 7:34 PM on February 15, 2023 [7 favorites]
Hell, I can go outside attach random meanings to various nuts thrown on my lawn and get an idea from which ones the squirrels leave for last, that does not make squirrels creative in any way shape or form.
posted by oddman at 7:34 PM on February 15, 2023 [7 favorites]
> "They often come up with ideas that I would call creative, if a human came up with those same ideas."
They don't "come up with ideas". Ever, at all. They do not have ideas, nor can they "come up with" anything.
You can certainly look at a pattern of frost on your window and decide it is as beautiful as any painting. But if you then praise the cold weather for being very creative in coming up with the idea for that pattern, you have missed something key.
posted by kyrademon at 7:48 PM on February 15, 2023 [34 favorites]
They don't "come up with ideas". Ever, at all. They do not have ideas, nor can they "come up with" anything.
You can certainly look at a pattern of frost on your window and decide it is as beautiful as any painting. But if you then praise the cold weather for being very creative in coming up with the idea for that pattern, you have missed something key.
posted by kyrademon at 7:48 PM on February 15, 2023 [34 favorites]
What is the difference between real AI and a computer just using tricks to seem intelligent -- fake AI?
Once we figure out how to get a computer to do something, we call it tricks -- just fake AI.
These new tricks -- somewhat coherent art from text requests, and ridiculously amazing chat bot tech -- may be a climax, or just the beginning of a very strange few decades.
posted by NotAYakk at 8:12 PM on February 15, 2023 [2 favorites]
Once we figure out how to get a computer to do something, we call it tricks -- just fake AI.
These new tricks -- somewhat coherent art from text requests, and ridiculously amazing chat bot tech -- may be a climax, or just the beginning of a very strange few decades.
posted by NotAYakk at 8:12 PM on February 15, 2023 [2 favorites]
"That said, I can see a real use for AI in tech writing. [...] But fiction? No. Freaking. Way."
Yeah, a shudder has run repeatedly down my tech writer spine in some of these AI discussions. The output won't be good in the part of the field that I worked in (software, sometimes as sole writer) but it'll pass muster until the brown and stinky hits the rotating blades on something important.
posted by gentlyepigrams at 8:26 PM on February 15, 2023 [2 favorites]
Yeah, a shudder has run repeatedly down my tech writer spine in some of these AI discussions. The output won't be good in the part of the field that I worked in (software, sometimes as sole writer) but it'll pass muster until the brown and stinky hits the rotating blades on something important.
posted by gentlyepigrams at 8:26 PM on February 15, 2023 [2 favorites]
too lazy to think or type. I asked chatgpt to write a metafilter comment for me:
AI-generated science fiction can be an interesting glimpse into what the future of machine-generated storytelling might hold. However, as with any automated writing, the quality of the output can be highly variable. While some AI-generated sci-fi stories may be surprisingly imaginative and thought-provoking, others may be nonsensical or derivative. Additionally, some readers may be put off by the idea of "robotic" storytelling that lacks the nuance and emotional depth of human-written fiction.
That being said, AI-generated science fiction can still be a valuable tool for exploring new ideas and generating inspiration for human writers. By analyzing patterns in existing science fiction literature, AI algorithms can help identify new plot points, themes, and character archetypes that human writers may not have thought of on their own. In this way, AI-generated science fiction can be a powerful tool for expanding the horizons of science fiction and inspiring new works of literature
posted by picklenickle at 8:29 PM on February 15, 2023 [7 favorites]
AI-generated science fiction can be an interesting glimpse into what the future of machine-generated storytelling might hold. However, as with any automated writing, the quality of the output can be highly variable. While some AI-generated sci-fi stories may be surprisingly imaginative and thought-provoking, others may be nonsensical or derivative. Additionally, some readers may be put off by the idea of "robotic" storytelling that lacks the nuance and emotional depth of human-written fiction.
That being said, AI-generated science fiction can still be a valuable tool for exploring new ideas and generating inspiration for human writers. By analyzing patterns in existing science fiction literature, AI algorithms can help identify new plot points, themes, and character archetypes that human writers may not have thought of on their own. In this way, AI-generated science fiction can be a powerful tool for expanding the horizons of science fiction and inspiring new works of literature
posted by picklenickle at 8:29 PM on February 15, 2023 [7 favorites]
I’m thinking of the day in the not-too-distant future when people will be able to have augmented/implanted chips that permit them to run software and feed them data, text, etc. in such a way that it’s not immediately obvious that they’re receiving the input.
Maybe it’s just an earphone-type device that connects externally and transmits sound through the bones in their skull. Maybe it’s something like Google Glass but more integrated. You know this will eventually happen.
As a college prof, I’ve been lamenting the day that happens, because proctoring in-person exams will become much more challenging. But imagine someone has access to a third- or fourth-generation ChatGPT app responding to their sub audible prompts.
You could be having a conversation with someone who is just parroting what their A.I. is feeding them in real-time. And you’d never know how much of your interaction with them, or how much they seem to know, or how much of their personality, is actually them or the A.I.
“Generate a informed response to the question we just heard. Include a reference to Degas, with an emphasis on his use of color. Conclude with a humorous but suave play on words related to Impressionism."
We won’t need to build robots that can think, because some people will essentially become the robots, happily providing the A.I. with its host body.
posted by darkstar at 8:29 PM on February 15, 2023 [13 favorites]
Maybe it’s just an earphone-type device that connects externally and transmits sound through the bones in their skull. Maybe it’s something like Google Glass but more integrated. You know this will eventually happen.
As a college prof, I’ve been lamenting the day that happens, because proctoring in-person exams will become much more challenging. But imagine someone has access to a third- or fourth-generation ChatGPT app responding to their sub audible prompts.
You could be having a conversation with someone who is just parroting what their A.I. is feeding them in real-time. And you’d never know how much of your interaction with them, or how much they seem to know, or how much of their personality, is actually them or the A.I.
“Generate a informed response to the question we just heard. Include a reference to Degas, with an emphasis on his use of color. Conclude with a humorous but suave play on words related to Impressionism."
We won’t need to build robots that can think, because some people will essentially become the robots, happily providing the A.I. with its host body.
posted by darkstar at 8:29 PM on February 15, 2023 [13 favorites]
What a pathetic metal horse!
The thing about horses is they think. The rider thinks. Model T can't think but the driver can. Somewere in the transition, a brain was taken out of the mix leaving for a brief period the illusion cars were somehow alive.
posted by clavdivs at 8:58 PM on February 15, 2023 [8 favorites]
The thing about horses is they think. The rider thinks. Model T can't think but the driver can. Somewere in the transition, a brain was taken out of the mix leaving for a brief period the illusion cars were somehow alive.
posted by clavdivs at 8:58 PM on February 15, 2023 [8 favorites]
picklenickle, if that's the response the AI gave you, I have to say I recognized it right away. It's the same wishywashy "Side A says this, while Side B says this, and in conclusion, it is all a land of contrasts" bunk that I'm trying to un-teach my junior high school students because it is largely terrible, devoid of anything that isn't already essentially common knowledge, and manages to take up space without really saying anything. From the structure to the amateurish transition words, it might as well have a neon sign saying "gotcha, you wasted your time reading this, and are no better off than if you hadn't."
It's something I'd hand back to my 7th graders with a page full of comments on how they could, instead, write something interesting that the reader might gain something from.
unless that's a stunt post, and you wrote it, in which case, hah, good one! You got me, and now let me go on about Pierre Menard, author of the Quixote!"
posted by Ghidorah at 9:10 PM on February 15, 2023 [23 favorites]
It's something I'd hand back to my 7th graders with a page full of comments on how they could, instead, write something interesting that the reader might gain something from.
unless that's a stunt post, and you wrote it, in which case, hah, good one! You got me, and now let me go on about Pierre Menard, author of the Quixote!"
posted by Ghidorah at 9:10 PM on February 15, 2023 [23 favorites]
Cut-up technique - Wikipedia
The cut-up technique (or découpé in French) is an aleatory literary technique in which a written text is cut up and rearranged to create a new text. The concept can be traced to the Dadaists of the 1920s, but it was developed and popularized in the 1950s and early 1960s, especially by writer William S. Burroughs. It has since been used in a wide variety of contexts.posted by ChurchHatesTucker at 9:36 PM on February 15, 2023 [5 favorites]
There is a certain pacing and use of filler phrases like "that being said" which seems to typify ChatGPT output. Hmmm.
posted by i_am_joe's_spleen at 9:37 PM on February 15, 2023 [5 favorites]
posted by i_am_joe's_spleen at 9:37 PM on February 15, 2023 [5 favorites]
The only valid use AI generated text has is generating the inexplicable 3000 word essay at the beginning of every online recipe.
posted by Reyturner at 9:38 PM on February 15, 2023 [20 favorites]
posted by Reyturner at 9:38 PM on February 15, 2023 [20 favorites]
But I mention this because today they emailed me to suggest I take a course called "Applied AI: Getting Started with Hugging Face Transformers."
That's all we need, an AI bursting out of our chests.
posted by gtrwolf at 10:10 PM on February 15, 2023 [1 favorite]
That's all we need, an AI bursting out of our chests.
posted by gtrwolf at 10:10 PM on February 15, 2023 [1 favorite]
Maybe it’s just an earphone-type device that connects externally and transmits sound through the bones in their skull. Maybe it’s something like Google Glass but more integrated. You know this will eventually happen.
o-or vibrating anal beads
posted by chavenet at 2:33 AM on February 16, 2023 [2 favorites]
o-or vibrating anal beads
posted by chavenet at 2:33 AM on February 16, 2023 [2 favorites]
By analyzing patterns in existing science fiction literature, AI algorithms can help identify new plot points, themes, and character archetypes that human writers may not have thought of on their own.
I love this line so much because if you spend any time at all with writer-focused AI, you realize just how untrue it is--it's a cheerfully optimistic statement so utterly at odds with reality that it's practically a mental illness model.
The one thing this technology cannot at all do is identify the new. It can't study the gaps. It doesn't even know the gaps are there. If you do ask it about plot points, it can summarize roughly what you could get from TV Tropes, I guess--it can tell you the blandest possible distillation of the blandest possible internet discussions of plot points, themes, and archetypes. In fact I used it recently for this, at a moment of panic with my current book--basically, sweet Jesus, please give me ten possible things that can go wrong for a detective at this part of a murder mystery, before I throw this story in the trash, and it dutifully gave me a listicle that saved me a google search. (Picking up ten mysteries and reading through them has been a longer, slower, but much more fun way to answer that question.)
AI differs from dice, I think, because dice are so constricted in what they can say, and you accept that going in. Inspiration comes from your acceptance of the limited, gnomic response. Even the mysterious and elusive plot wheels of our pulp forebears offer very little, forcing the brain to do its work. But because AI seeks to be more complete than that, the weaknesses in the system show.
posted by mittens at 3:21 AM on February 16, 2023 [13 favorites]
I love this line so much because if you spend any time at all with writer-focused AI, you realize just how untrue it is--it's a cheerfully optimistic statement so utterly at odds with reality that it's practically a mental illness model.
The one thing this technology cannot at all do is identify the new. It can't study the gaps. It doesn't even know the gaps are there. If you do ask it about plot points, it can summarize roughly what you could get from TV Tropes, I guess--it can tell you the blandest possible distillation of the blandest possible internet discussions of plot points, themes, and archetypes. In fact I used it recently for this, at a moment of panic with my current book--basically, sweet Jesus, please give me ten possible things that can go wrong for a detective at this part of a murder mystery, before I throw this story in the trash, and it dutifully gave me a listicle that saved me a google search. (Picking up ten mysteries and reading through them has been a longer, slower, but much more fun way to answer that question.)
AI differs from dice, I think, because dice are so constricted in what they can say, and you accept that going in. Inspiration comes from your acceptance of the limited, gnomic response. Even the mysterious and elusive plot wheels of our pulp forebears offer very little, forcing the brain to do its work. But because AI seeks to be more complete than that, the weaknesses in the system show.
posted by mittens at 3:21 AM on February 16, 2023 [13 favorites]
AI differs from dice also because AI output depends on the previous roll, so to speak. Dice do not.
posted by JohnFromGR at 4:07 AM on February 16, 2023 [4 favorites]
posted by JohnFromGR at 4:07 AM on February 16, 2023 [4 favorites]
I wonder if anyone submitted a story to, say, Analog in 1983 dealing with the headaches faced by an sf magazine editor who gets too many manuscripts authored by computers.
posted by ricochet biscuit at 4:40 AM on February 16, 2023 [14 favorites]
posted by ricochet biscuit at 4:40 AM on February 16, 2023 [14 favorites]
Any sufficiently advanced dice rolls...
posted by Jacen at 5:32 AM on February 16, 2023 [4 favorites]
posted by Jacen at 5:32 AM on February 16, 2023 [4 favorites]
I've seen other folks attempt to prod ChatGPT into writing a story as if by me. (I don't have a ChatGPT account; can't be bothered.) It was truly dire.
The thing is, ChatGPT is a probabilistic system for generating prose iteratively. Each sentence is well-formed and follows on from the previous one. But there are no long range interconnections—good luck expecting it to pull the trigger in act three on the gun it placed on the mantelpiece in act one. The sort of output we're seeing from GANs that are marketed as AI this decade show no sign of object permanence. (There were programs that could handle object permanence within a restricted realm back in the 1960s (such as Terry Winograd's SHRDLU) but today's GPT models don't have it, they're something different.)
Now, there are uses for ChatGPT in fiction. I've seen what results when you feed a film script to ChatGPT and tell it to rewrite it as a novelization. It's not a publication-grade novelization because the script is missing out a whole lot of stuff (actors' facial expressions, scenery, smell, lighting, descriptive texture) but it's a creditable first pass which could make a media tie-in author's job much easier by at least getting the dialogue right.
Similarly, you can feed a novel into ChatGPT and tell it to "rewrite this as a film script" and ... it's not a film script, but it's a film-script-shaped-object that a human screenwriter can then edit.
Now, the whole point of Clarkesworld is that readers are there for the unexpected, for the human creative element. And ChatGPT can't deliver that, except accidentally, because it's extrapolating from a repertoire of prior art and the unexpected is by definition not part of the body of prior art.
As to where these spam submissions are coming from: Magazines like Clarkesworld pay their contributors (the successful published ones, that is). The pay is peanuts but it's real money. I suspect the people who send these auto-generated spam submissions (who I also suspect are not readers of fiction in general and Clarkesworld in particular) are used to earning income by generating text for content mills that publish everything uncritically, without editorial supervision (because the content mills are actually link farming for ad revenue). ChatGPT is destroying their market, so they're looking for new pastures to colonize. And at Clarkesworld they've run into a highly competent human editor who says "nope".
It's as simple as that.
posted by cstross at 5:44 AM on February 16, 2023 [57 favorites]
The thing is, ChatGPT is a probabilistic system for generating prose iteratively. Each sentence is well-formed and follows on from the previous one. But there are no long range interconnections—good luck expecting it to pull the trigger in act three on the gun it placed on the mantelpiece in act one. The sort of output we're seeing from GANs that are marketed as AI this decade show no sign of object permanence. (There were programs that could handle object permanence within a restricted realm back in the 1960s (such as Terry Winograd's SHRDLU) but today's GPT models don't have it, they're something different.)
Now, there are uses for ChatGPT in fiction. I've seen what results when you feed a film script to ChatGPT and tell it to rewrite it as a novelization. It's not a publication-grade novelization because the script is missing out a whole lot of stuff (actors' facial expressions, scenery, smell, lighting, descriptive texture) but it's a creditable first pass which could make a media tie-in author's job much easier by at least getting the dialogue right.
Similarly, you can feed a novel into ChatGPT and tell it to "rewrite this as a film script" and ... it's not a film script, but it's a film-script-shaped-object that a human screenwriter can then edit.
Now, the whole point of Clarkesworld is that readers are there for the unexpected, for the human creative element. And ChatGPT can't deliver that, except accidentally, because it's extrapolating from a repertoire of prior art and the unexpected is by definition not part of the body of prior art.
As to where these spam submissions are coming from: Magazines like Clarkesworld pay their contributors (the successful published ones, that is). The pay is peanuts but it's real money. I suspect the people who send these auto-generated spam submissions (who I also suspect are not readers of fiction in general and Clarkesworld in particular) are used to earning income by generating text for content mills that publish everything uncritically, without editorial supervision (because the content mills are actually link farming for ad revenue). ChatGPT is destroying their market, so they're looking for new pastures to colonize. And at Clarkesworld they've run into a highly competent human editor who says "nope".
It's as simple as that.
posted by cstross at 5:44 AM on February 16, 2023 [57 favorites]
Wow posting after cstross... no original idea here but I kinda liked a snipit from the article commentsection by "TJ"
I think they are using a kind of litigious magical thinking about what constitutes a submission, and playing into a Post-Marxist (yikes that phrase) caricature of commodity fetishism where a story is whatever you can get paid for,
posted by sammyo at 6:16 AM on February 16, 2023 [1 favorite]
I think they are using a kind of litigious magical thinking about what constitutes a submission, and playing into a Post-Marxist (yikes that phrase) caricature of commodity fetishism where a story is whatever you can get paid for,
posted by sammyo at 6:16 AM on February 16, 2023 [1 favorite]
Yesterday, I was in a Zoom meeting as part of a career transitions program I am participating in. A person who is also in the program went on a ten minute tangent about ChatGPT and how it might hinder and/or help them in their future job prospects. This AI stuff is just posted everywhere these last couple months and it really is getting more exposure than it probably warrants.
However, I enjoy some fantasy fiction now and then. I'll read about some specific novel recommended somewhere then go online to find out more about it or potentially buy it for my e-reader. The sheer gargantuan amount of fantasy fiction available at the click of a finger these days, self-published and otherwise, suggests to me that an automated algorithm could probably indeed produce a marketable and polished (enough) novel as long as someone slaps a picture of a dragon on the cover. Does this mean I need to find out ways to avoid a ChatGPT-created fantasy novel? No, it just means I need to keep on being discriminating in what I spend money on and read.
posted by SoberHighland at 6:48 AM on February 16, 2023
However, I enjoy some fantasy fiction now and then. I'll read about some specific novel recommended somewhere then go online to find out more about it or potentially buy it for my e-reader. The sheer gargantuan amount of fantasy fiction available at the click of a finger these days, self-published and otherwise, suggests to me that an automated algorithm could probably indeed produce a marketable and polished (enough) novel as long as someone slaps a picture of a dragon on the cover. Does this mean I need to find out ways to avoid a ChatGPT-created fantasy novel? No, it just means I need to keep on being discriminating in what I spend money on and read.
posted by SoberHighland at 6:48 AM on February 16, 2023
A chatgpt-generated fantasy novel will also probably have a stable diffusion-generated dragon on the cover, clutching an ambiguous shiny object in its seventeen-fingered claws.
posted by egypturnash at 6:54 AM on February 16, 2023 [5 favorites]
posted by egypturnash at 6:54 AM on February 16, 2023 [5 favorites]
I'm in the middle of starting an online magazine for SFF flash fiction and poetry with a friend of mine, and man, is this hitting hard. Online SFF short fiction magazines like Clarkesworld have first readers who filter through what's submitted. It's a hard job that can burn you out. Flooding the submission pipeline with ML-written dreck is going to burn first readers out much faster.
Neil mentions the shakiness of automated detection solutions. If they get much better, what do you want to bet they'll be priced beyond what a small magazine limping along can afford? Limited submission windows might help, but how long before some bright spark combines a tool that tracks which magazines are open for submissions like The Submission Grinder with automated submissions?
The solution may kill SFF magazines' (relative) openness to newcomers. You can throttle the ML submissions by charging a submission fee, like lit magazines do, or getting rid of open submissions and only taking invited stories, like The Sunday Morning Transport.
Both of those answers make me tired and sad. I grew up reading short stories. I love the form enough to launch a new magazine into the headwinds of changes like Amazon nuking the Kindle Publishing for Periodicals program for small magazines. I want the short fiction market to be welcoming to newcomers and open to experimentation. These recent developments push against those goals.
We'll figure out a way, but it increasingly feels like the deck is not only stacked against small magazines but is also on fire.
posted by sgranade at 6:57 AM on February 16, 2023 [11 favorites]
Neil mentions the shakiness of automated detection solutions. If they get much better, what do you want to bet they'll be priced beyond what a small magazine limping along can afford? Limited submission windows might help, but how long before some bright spark combines a tool that tracks which magazines are open for submissions like The Submission Grinder with automated submissions?
The solution may kill SFF magazines' (relative) openness to newcomers. You can throttle the ML submissions by charging a submission fee, like lit magazines do, or getting rid of open submissions and only taking invited stories, like The Sunday Morning Transport.
Both of those answers make me tired and sad. I grew up reading short stories. I love the form enough to launch a new magazine into the headwinds of changes like Amazon nuking the Kindle Publishing for Periodicals program for small magazines. I want the short fiction market to be welcoming to newcomers and open to experimentation. These recent developments push against those goals.
We'll figure out a way, but it increasingly feels like the deck is not only stacked against small magazines but is also on fire.
posted by sgranade at 6:57 AM on February 16, 2023 [11 favorites]
AI differs from dice also because AI output depends on the previous roll, so to speak. Dice do not.
They can. Dice rolls are only meaningful because of the meaning we assign to outcomes ahead of rolling. Role playing games that use dice often have result tables that are dependent on the results of previous rolls. EG: you are generating the contents of chest. First you roll a d10 to determine how many objects are in the chest. Then you roll a d20 to determine what sort (weapon/potion/scroll/coins/gems/artifacts/armor/lore/implement/etc.) each of the objects is. Then to take weapon for example you could roll a d20 to see if it is magical or not. Say it turns up magical you could roll d100 to see what class of weapon it is (pole arm/sword/bow/mace/etc.) And then finally roll a d100 to determine exactly what sword the player found in the chest (+1 vorpal dagger of striking).
And each of those rolls can be weighted. The reason you roll a d20 for a binary choice instead of a d2 is 1-19 results are all mundane and only a 20 gets you a magical result. The same with the d100 roll for weapon class. If 30% of found weapons are daggers you represent that in the look up table with thirty results indicating dagger. Or maybe you use a bell curve generated by rolling multiple dice and adding them together. You could roll 3d6 and assign each of the 15 results to a weapon type and whatever is assigned 10 is going to be picked way more often that the 3 result.
posted by Mitheral at 7:38 AM on February 16, 2023 [1 favorite]
They can. Dice rolls are only meaningful because of the meaning we assign to outcomes ahead of rolling. Role playing games that use dice often have result tables that are dependent on the results of previous rolls. EG: you are generating the contents of chest. First you roll a d10 to determine how many objects are in the chest. Then you roll a d20 to determine what sort (weapon/potion/scroll/coins/gems/artifacts/armor/lore/implement/etc.) each of the objects is. Then to take weapon for example you could roll a d20 to see if it is magical or not. Say it turns up magical you could roll d100 to see what class of weapon it is (pole arm/sword/bow/mace/etc.) And then finally roll a d100 to determine exactly what sword the player found in the chest (+1 vorpal dagger of striking).
And each of those rolls can be weighted. The reason you roll a d20 for a binary choice instead of a d2 is 1-19 results are all mundane and only a 20 gets you a magical result. The same with the d100 roll for weapon class. If 30% of found weapons are daggers you represent that in the look up table with thirty results indicating dagger. Or maybe you use a bell curve generated by rolling multiple dice and adding them together. You could roll 3d6 and assign each of the 15 results to a weapon type and whatever is assigned 10 is going to be picked way more often that the 3 result.
posted by Mitheral at 7:38 AM on February 16, 2023 [1 favorite]
One solution could be the good old mefi strategy of requiring $5 to submit. Might be enough to keep out (most of) the spammers.
posted by dhruva at 7:03 PM on February 1
Probably, but it has real problems: Clarksworld, like a lot of SFF markets, have a catchment area of the planet, and any submission fee is going to keep out, e.g. the amazing writer from a low-resource country for whom $5 US is a month's wages or whatever.
Lit mags went the route of asking for submissions fees a few decades ago and the field has gotten significantly less diverse from a class perspective, at least in my opinion (while it's gotten more diverse from race, gender, and sexuality perspectives, they still tend to be the voices of people who can afford to get an MFA).
I'd hate to see SFF mags go that route. However, I'm not sure how to work around this sort of thing otherwise. Can ChatGPT be taught to distinguish between AI-generated and real-human-generated stories and act as a first reader?
posted by joannemerriam at 9:05 AM on February 16, 2023 [2 favorites]
posted by dhruva at 7:03 PM on February 1
Probably, but it has real problems: Clarksworld, like a lot of SFF markets, have a catchment area of the planet, and any submission fee is going to keep out, e.g. the amazing writer from a low-resource country for whom $5 US is a month's wages or whatever.
Lit mags went the route of asking for submissions fees a few decades ago and the field has gotten significantly less diverse from a class perspective, at least in my opinion (while it's gotten more diverse from race, gender, and sexuality perspectives, they still tend to be the voices of people who can afford to get an MFA).
I'd hate to see SFF mags go that route. However, I'm not sure how to work around this sort of thing otherwise. Can ChatGPT be taught to distinguish between AI-generated and real-human-generated stories and act as a first reader?
posted by joannemerriam at 9:05 AM on February 16, 2023 [2 favorites]
There's this GPT-2 output detector which seems quite good, and ranks picklenickle's AI-generated comment as 99.98% likely to be fake.
posted by heatherlogan at 9:10 AM on February 16, 2023 [6 favorites]
posted by heatherlogan at 9:10 AM on February 16, 2023 [6 favorites]
This is all not conscious, but just a giant lookup table. But that means it's a tool, like divination is a tool. A chart of meaning, a syntactic tree
I think there is something interesting about machines producing human-like output, but unconsciously. Ironically it kind of makes me want to read an SF story about someone discovering that god exists, is omniscient, omnipotent, but has no consciousness whatsoever. Which, thinking about it, seems weirdly acceptable to me.
posted by Jon Mitchell at 9:25 AM on February 16, 2023 [6 favorites]
I think there is something interesting about machines producing human-like output, but unconsciously. Ironically it kind of makes me want to read an SF story about someone discovering that god exists, is omniscient, omnipotent, but has no consciousness whatsoever. Which, thinking about it, seems weirdly acceptable to me.
posted by Jon Mitchell at 9:25 AM on February 16, 2023 [6 favorites]
AI-generated spam might be the straw the breaks the RFC-822 camel's back. And provides the impetus, finally, for something like X.500 being rolled out to the whole human population. Which will be its own kind of nightmare, and won't totally exclude the bots.
posted by Aardvark Cheeselog at 9:33 AM on February 16, 2023 [1 favorite]
posted by Aardvark Cheeselog at 9:33 AM on February 16, 2023 [1 favorite]
When the Singularity comes, how will we know it's not just some badly written AI-generated fiction?
posted by Greg_Ace at 10:00 AM on February 16, 2023 [2 favorites]
posted by Greg_Ace at 10:00 AM on February 16, 2023 [2 favorites]
Greg_Ace: I seem to remember back in 2010 publishing a novel which introduced the concept of the Spamularity:
Oh, and because no idea is original, I was riffing off an inspired rant/lecture by Cory Doctorow from about 2007. Once you start talking Alan Turing's paper that introduced The Imitation Game seriously (rather than knocking it for his ghastly, antiquated gender hangups) it's all present by implication.
posted by cstross at 10:34 AM on February 16, 2023 [9 favorites]
Today, in the wake of the bursting of the worldwide higher education bubble, the big funding sources in computational artificial intelligence research are computer games and cognitive marketing services, from personalised message generation to automated spear phishing. Some say the spammers are pouring more money into Minsky's inheritors than the US Department of Defence ever imagined. The spamcops retaliate. There's an arms race in progress, and some experts mutter dark warnings of the Spamularity: the global communications gridlock that will ensue once the first distributed spamming engine achieves human-equivalent sentience.I am not pleased to have called it right nearly 15 years ago (I wrote Rule 34 in 2008-09).
Oh, and because no idea is original, I was riffing off an inspired rant/lecture by Cory Doctorow from about 2007. Once you start talking Alan Turing's paper that introduced The Imitation Game seriously (rather than knocking it for his ghastly, antiquated gender hangups) it's all present by implication.
posted by cstross at 10:34 AM on February 16, 2023 [9 favorites]
This is also one of the clever ideas in Peter Watts's "Rifters" trilogy. Published 1999-2004, but I don't remember which book is the one that explains that things like email are all but unusable because of the AI spam arms race. He does propose that a temporary advantage has been won by having cultured human brain tissue gatekeepers installed along the infrastructure, which I guess in a narrow way is a bit optimistic (readers of Watts will of course know that it is not optimistic).
posted by grobstein at 12:14 PM on February 16, 2023
posted by grobstein at 12:14 PM on February 16, 2023
As usual for every "original" joke I come up with, it turns out someone else got there first. *sigh*
:)
posted by Greg_Ace at 12:24 PM on February 16, 2023
:)
posted by Greg_Ace at 12:24 PM on February 16, 2023
(Also skipping in from 2007) PGP networks of trust will fix this!
posted by clew at 1:09 PM on February 16, 2023 [3 favorites]
posted by clew at 1:09 PM on February 16, 2023 [3 favorites]
Anybody whose big idea is getting rich by gaming the short fiction market is a stone cold fucking idiot.
posted by kittens for breakfast at 3:32 PM on February 16, 2023 [17 favorites]
posted by kittens for breakfast at 3:32 PM on February 16, 2023 [17 favorites]
Can I reinforce my whole-hearted agreement to Ghidorah's comment.
My teenagers are able to generate verbiage at a torrential speed - which is pretty much indistinguishable from most of the ChatGPT output. If I force them to use pen and paper, then the linearity and physical effort seems to improve the quality, but there is no such constraint on these ML/AI processes.
With ML/AI, the syntax is fine - but the semantics are non-existent. I use words to structure a concept that another person can recognise as that concept. Even if the words are forgotten, the concept may remain. These floods of words are purposeless, so there is no structure, and there is no concept.
posted by Barbara Spitzer at 5:44 PM on February 16, 2023 [4 favorites]
My teenagers are able to generate verbiage at a torrential speed - which is pretty much indistinguishable from most of the ChatGPT output. If I force them to use pen and paper, then the linearity and physical effort seems to improve the quality, but there is no such constraint on these ML/AI processes.
With ML/AI, the syntax is fine - but the semantics are non-existent. I use words to structure a concept that another person can recognise as that concept. Even if the words are forgotten, the concept may remain. These floods of words are purposeless, so there is no structure, and there is no concept.
posted by Barbara Spitzer at 5:44 PM on February 16, 2023 [4 favorites]
If you take a bootstrap sample (from n data points, randomly select n of them, with replacement) then on average you expect something like 1-1/e of the original n data points to be in your sample.
In a broad stroke, it seems like we're heading toward these AI approaches taking the whole entire internet as a training set, and then (with our help) regurgitating a new (bootstrap sample?) internet in parallel.
In such a case, the part that would seem redundant to a well-read observer would be that 1/e that got repeated. Which is creepily close to the 38% of spammy submissions that got banned.
*also, <3 for "sparkling autocomplete." This will be the name of my AI-produced greatest hits album.
posted by adekllny at 6:13 PM on February 16, 2023
In a broad stroke, it seems like we're heading toward these AI approaches taking the whole entire internet as a training set, and then (with our help) regurgitating a new (bootstrap sample?) internet in parallel.
In such a case, the part that would seem redundant to a well-read observer would be that 1/e that got repeated. Which is creepily close to the 38% of spammy submissions that got banned.
*also, <3 for "sparkling autocomplete." This will be the name of my AI-produced greatest hits album.
posted by adekllny at 6:13 PM on February 16, 2023
Anybody whose big idea is getting rich by gaming the short fiction market is a stone cold fucking idiot.
Anybody whose big idea is getting rich by sending sob stories about a Nigerian prince to every email address on the planet is a stone cold idiot.
posted by straight at 10:41 PM on February 16, 2023 [2 favorites]
Anybody whose big idea is getting rich by sending sob stories about a Nigerian prince to every email address on the planet is a stone cold idiot.
posted by straight at 10:41 PM on February 16, 2023 [2 favorites]
Trying to remember where I read that one of the larger datasets for training ML/AI was the Enron emails - right up there with Nigerian princes for a narrative database.
posted by Barbara Spitzer at 11:32 PM on February 16, 2023
posted by Barbara Spitzer at 11:32 PM on February 16, 2023
I feel like there's something entertaining, if only for a few minutes, by training the LLMs on limited corpuses. The complete scripts of the Brady Bunch or the Saturday morning cartoons of the 70-80s. Pink Floyd lyrics. Google Translate-and-back of the Bible or Finnegan's Wake. What would robot-Roger Waters say I should do with this bag of shit I'm about to smear on a critic toward whom I feel much antipathy?
posted by rhizome at 11:55 PM on February 16, 2023 [1 favorite]
posted by rhizome at 11:55 PM on February 16, 2023 [1 favorite]
In a broad stroke, it seems like we're heading toward these AI approaches taking the whole entire internet as a training set
I'd be curious to see how one that ate JSTOR would be different.
posted by GCU Sweet and Full of Grace at 5:12 AM on February 17, 2023 [4 favorites]
I'd be curious to see how one that ate JSTOR would be different.
posted by GCU Sweet and Full of Grace at 5:12 AM on February 17, 2023 [4 favorites]
Anybody whose big idea is getting rich by gaming the short fiction market is a stone cold fucking idiot.
That's one of the things that makes me especially angry about this topic. I'm a science fiction writer, still trying to get published, and I've been chugging away on a steady stream of stories and rejections for a while now. I know that even when I do sell a story, weeks of work and months of the submit-wait-rejection cycle will result in a paycheck of about $400 for a 5000-word story, and that's probably all I'll ever make from that story. And that's if I'm lucky and able to get the "pro" rate of $0.08 per word. My fellow writer friends already play submission tetris, because the industry standard for short stories is that you can only send them to one place at a time. Sometimes it takes a long time to hear back, on the order of multiple months. In the meantime, other places you might want to send the story finally open their submission windows, which may only be open, say, twice a year, for two weeks at a time. If your story is still in someone else's slush pile when the blue moon happens, oh well. Try again the next time the stars align.
Clarkesworld, on the other hand, is special. They pay $0.12 a word. Their average turnaround time is three days. They're a wonderful first place to send a piece because they're always open, they don't limit themselves to soliciting the same names over and over, and you'll get it back fast. They publish lots of new voices because their writer-friendly practices mean they get first crack at a lot of stories. But, even if the editors were to flood their magazine with the AI-generated garbage that spammers are dumping on them now, any given spammer will still only make $600. That's not nothing, and I'd love to get that much for a story, but AI dreck is not going to get published in any reasonable magazine.
All they're doing is degrading one of the best markets for short speculative fiction. The first readers and editors on the other side are human (as far as I know). They'll burn out. If the deluge continues, they'll have to change something, and magazine publishing is already pretty tough financially--and it's not getting better. I hate that lazy spammers, who aren't going to make any money anyway, are making it even tougher to run a fiction magazine. Selfishly, I hate that it'll become harder for me to get my stories published as a nobody, as publications have to increase barriers or slow down in some way.
Throw it on the pile of ways neural nets make the internet, and trying to have a creative career, just a little more garbage.
posted by j.r at 9:46 AM on February 17, 2023 [13 favorites]
That's one of the things that makes me especially angry about this topic. I'm a science fiction writer, still trying to get published, and I've been chugging away on a steady stream of stories and rejections for a while now. I know that even when I do sell a story, weeks of work and months of the submit-wait-rejection cycle will result in a paycheck of about $400 for a 5000-word story, and that's probably all I'll ever make from that story. And that's if I'm lucky and able to get the "pro" rate of $0.08 per word. My fellow writer friends already play submission tetris, because the industry standard for short stories is that you can only send them to one place at a time. Sometimes it takes a long time to hear back, on the order of multiple months. In the meantime, other places you might want to send the story finally open their submission windows, which may only be open, say, twice a year, for two weeks at a time. If your story is still in someone else's slush pile when the blue moon happens, oh well. Try again the next time the stars align.
Clarkesworld, on the other hand, is special. They pay $0.12 a word. Their average turnaround time is three days. They're a wonderful first place to send a piece because they're always open, they don't limit themselves to soliciting the same names over and over, and you'll get it back fast. They publish lots of new voices because their writer-friendly practices mean they get first crack at a lot of stories. But, even if the editors were to flood their magazine with the AI-generated garbage that spammers are dumping on them now, any given spammer will still only make $600. That's not nothing, and I'd love to get that much for a story, but AI dreck is not going to get published in any reasonable magazine.
All they're doing is degrading one of the best markets for short speculative fiction. The first readers and editors on the other side are human (as far as I know). They'll burn out. If the deluge continues, they'll have to change something, and magazine publishing is already pretty tough financially--and it's not getting better. I hate that lazy spammers, who aren't going to make any money anyway, are making it even tougher to run a fiction magazine. Selfishly, I hate that it'll become harder for me to get my stories published as a nobody, as publications have to increase barriers or slow down in some way.
Throw it on the pile of ways neural nets make the internet, and trying to have a creative career, just a little more garbage.
posted by j.r at 9:46 AM on February 17, 2023 [13 favorites]
it seems like we're heading toward these AI approaches taking the whole entire internet as a training set
I suspect the problem of recursion / repetition in the 570 GB GPT-3 dataset was a big hurdle, and the very existence of chatGPT et al spewing NLP-derived variants into the void is going to make the problem exponentially worse.
posted by aspersioncast at 1:53 PM on February 17, 2023 [1 favorite]
I suspect the problem of recursion / repetition in the 570 GB GPT-3 dataset was a big hurdle, and the very existence of chatGPT et al spewing NLP-derived variants into the void is going to make the problem exponentially worse.
posted by aspersioncast at 1:53 PM on February 17, 2023 [1 favorite]
I wonder if anyone submitted a story to, say, Analog in 1983 dealing with the headaches faced by an sf magazine editor who gets too many manuscripts authored by computers.
I sent the article and this comment to my neighbor who has published 104 stories in Analog since 1982. He says "I wouldn't be surprised if someone did, but if so I'm not familiar with the story. In general science fiction writers are pretty awful at predicting things."
posted by neuron at 2:33 PM on February 17, 2023 [4 favorites]
I sent the article and this comment to my neighbor who has published 104 stories in Analog since 1982. He says "I wouldn't be surprised if someone did, but if so I'm not familiar with the story. In general science fiction writers are pretty awful at predicting things."
posted by neuron at 2:33 PM on February 17, 2023 [4 favorites]
It's quite like a Roald Dahl short story: The Great Automatic Grammatizer
Although the prospects in that for the money-making aspects of such a machine are a bit more optimistic than they would be today I think.
posted by colourlesssleep at 5:27 AM on February 18, 2023 [2 favorites]
Although the prospects in that for the money-making aspects of such a machine are a bit more optimistic than they would be today I think.
posted by colourlesssleep at 5:27 AM on February 18, 2023 [2 favorites]
clarkesworld on Twitter:
Submissions are currently closed. It shouldn't be hard to guess why.
posted by thatwhichfalls at 1:54 PM on February 20, 2023 [3 favorites]
Submissions are currently closed. It shouldn't be hard to guess why.
posted by thatwhichfalls at 1:54 PM on February 20, 2023 [3 favorites]
We're all living in a Ted Chiang story.Hell Is The AI's Sense Of Good?
posted by I'm always feeling, Blue at 4:27 PM on February 20, 2023
And now, the NYT version.
"The writing is also “bad in spectacular ways,” Mr. Clarke said. “They’re just prompting, dumping, pasting and submitting to a magazine.”
Sheila Williams, the editor of Asimov’s Science Fiction magazine, said that several of the chatbot-generated stories she had received all had the same title: “The Last Hope.”
“The people doing this by and large don’t have any real concept of how to tell a story, and neither do any kind of A.I.,” Ms. Williams said on Wednesday. “You don’t have to finish the first sentence to know it’s not going to be a readable story.”
Ms. Thomas said that she had been permanently banning anyone who submitted chatbot-generated work.
“I don’t want to read bot stories,” she said. “I want to read stories that come out of actual imagination and experiences, and their own impulses.”
Scalzi also has some thoughts.
posted by jenfullmoon at 2:25 PM on February 23, 2023 [1 favorite]
"The writing is also “bad in spectacular ways,” Mr. Clarke said. “They’re just prompting, dumping, pasting and submitting to a magazine.”
Sheila Williams, the editor of Asimov’s Science Fiction magazine, said that several of the chatbot-generated stories she had received all had the same title: “The Last Hope.”
“The people doing this by and large don’t have any real concept of how to tell a story, and neither do any kind of A.I.,” Ms. Williams said on Wednesday. “You don’t have to finish the first sentence to know it’s not going to be a readable story.”
Ms. Thomas said that she had been permanently banning anyone who submitted chatbot-generated work.
“I don’t want to read bot stories,” she said. “I want to read stories that come out of actual imagination and experiences, and their own impulses.”
Scalzi also has some thoughts.
posted by jenfullmoon at 2:25 PM on February 23, 2023 [1 favorite]
« Older It Monocultures Your Thoughts | German ballet director suspended over poop attack... Newer »
This thread has been archived and is closed to new comments
"That said, I can see a real use for AI in tech writing. [...] But fiction? No. Freaking. Way."
Tech writers, of course, argue the reverse.
posted by kaibutsu at 4:25 PM on February 15, 2023 [24 favorites]