Sci-Fi Mag Pauses Submissions Amid Flood of AI-Generated Short Stories (pcmag.com) 71
The rise of AI-powered chatbots is wreaking havoc on the literary world. Sci-fi publication Clarkesworld Magazine is temporarily suspending short story submissions, citing a surge in people using AI chatbots to "plagiarize" their writing. From a report: The magazine announced the suspension days after Clarkesworld editor Neil Clarke warned about AI-written works posing a threat to the entire short-story ecosystem. At the end of last year, the sci-fi publication encountered a rise in plagiarism as AI-powered chatbots gained the public's attention, Clarke wrote in a blog post. Since then, Clarkesworld has seen a massive spike in short story submissions, but much of the writing appears to come from humans relying on AI tools to pump out the text.
They need to adjust their monetization policy (Score:1, Redundant)
surge in people using AI chatbots to "plagiarize" their writing.
Just bc they found a tool to help them doesn't make it plagiarism.. if the AI chatbots present a problem, then make a rule, and call it a rule violation.
since the sci-fi publication pays 12 cents a word for accepted stories.
Oh... so they've created an incentive with no cost for abuse (e.g. low quality automated garbage). That is the problem: fine if you pay for accepted stories, but it seems like you need an upfront submission fee that w
Re: (Score:3)
The current AI methods literally are copying text or images and modifying it. The AI does not come up with originality. They either have a huge existing database or they have internet access. So when a user says "write me a story about dog angels" then it'd go and search for lots of stories and paragraphs and plot summaries and mix it all up; thus you're very likely get nearly word for word copies of sentences and short paragraphs that are only slightly modified from the original.
Re: They need to adjust their monetization policy (Score:2)
That isn't how the image tools work (contrary to a billion articles that describe it wrong).
The text tools may or may not be different than described. I haven't been impressed with them, and often they do seem to clone text. However, much of that text has no copyright, making it even more complicated.
(Not every bit of text can be copyrighted stand alone.)
Re: They need to adjust their monetization policy (Score:4, Interesting)
Plagiarism of uncopyrighted material is still plagiarism though. And that matters to a site that accepts short stories.
Re: (Score:3)
I have personally seen a well-known AI translation web site feed my own translation back to me, word for word. AI anything is plagiarism.
Re: (Score:2)
"AI anything is plagiarism."
I think the only way you can support that assertion is by also declaring that "Any creative act by people is also plagiarism". After all, anything created by anybody currently alive (and probably for a great many dead generations) is in a very real sense a derivative work. Disney invented neither animation nor the mouse.
But we don't think that way, so the test is whether or not the work is significantly unique enough from other samples to constitute an original creation. And I don't know why the same test s
Re: (Score:2, Informative)
You AI fanbois have a terrible view on creativity. Humans do not just sample and regurgitate, but where they do we recognize that it requires attribution.
Re: (Score:3)
I think you're missing my point. I'm not particularly a fan of AI, nor a detractor.
My point is that people are trained on models too. Nobody writes a blues song without hearing blues songs. And there was no "blues origin story" - ragtime, church, and folk raised a baby. Do people just sample and regurgitate? No, they transform. But so do AI systems. If your decision on originality is based solely on, "One comes from people, the other doesn't", then we can't really have a discussion. If the objection is more
Re: (Score:1)
If your mom used to hum a melody when you were little, and later you became a musician and unconsciously came up with a melody extremely similar to the one your mom used to hum, without knowing that it was a famous song back when your mom was a teenager, then that's still a copyright violation. Generally this doesn't happen, because people tell you you're plagiarizing before it becomes a problem. But AIs do everything like that. They are not "inspired" by the works of artists to create their own. They liter
Re: (Score:1)
Re: (Score:2)
Re: (Score:2)
I wouldn't worry overly much about that. 26 letters and a handful of accessories built millions of books. Imagine what diversity there can be with 88 keys.
Re: (Score:2)
Re: (Score:2)
You missed my point. The vast majority of 26 letter combinations are nonsense too. Published works are only the coherent ones. Same for notes.
Re: (Score:2)
Actual plagiarism entails two harms: (1) denial of credit to the true author and (2) attribution of credit to the undeserving claimant. So what AI has done at the very least is enable new kind of "demi-plagiarism" in which someone can fraudulently claim credit for work he didn't do, but no other author is harmed.
Does it matter? Not much in isolation, but as a trend it's going to create big problems. AI in its current state of development does an impressive job of creating mediocre output -- good enough to
Re: (Score:2)
I have personally seen a well-known AI translation web site feed my own translation back to me, word for word. AI anything is plagiarism.
You don't have a sound argument there. It is the equivalent of "I once saw a Forum comment that was a word-for-word Copy and Paste of a Wikipedia article.
So all comments on internet forums are plagiarized"
Re: (Score:1)
That's not how LMs work, and it's not how diffusion models work.
Do you think overnight we've developed compression algorithms that condense 5 TB of JPEGs into 50 GB?
Re: (Score:2)
That's only a factor 100. Yes, it's lossy compression, but it's certainly not understanding and creating. It's mixing and matching. Call it a collage, if you want to, but it lacks all creativity. It's plagiarism.
Re: (Score:2)
There's nothing collage like about it. It's not "mixing and matching". It's certainly not compression!
You might want to learn something about the tech from somewhere that isn't a social media post or pop sci blog.
it lacks all creativity.
Yes.
It's plagiarism.
No.
Re: They need to adjust their monetization policy (Score:2)
Re: (Score:2)
The legality has yet to be determined.
Re:They need to adjust their monetization policy (Score:4, Interesting)
That's only a factor 100. Yes, it's lossy compression
"Only" a factor of 100.. LAION-5 was more than 200 TB of images; the trained network for SD was approximately 10 Gigabytes - 1 pixel per image would require more space than that; these systems are Not lossly-compressed archives of training data.
In some cases some of the training data could be reproduced - usually it is Not feasible to get the trained networks to reproduce one of their training inputs, short of prompting with it.
Re: (Score:2)
AIs recreate the parts that occur often in the training data. It's not lossy compression like a JPEG, but the model is an encoding of the "important" parts of the training data. What else do you think it is?
Re: (Score:1)
What you described isn't a collage and isn't copying and pasting. It's describing a process of noticing the important aspects of what creates a desirable output for a certain task, generalizing the methods by which to create it, and applying that across the problem domain. Sounds an awful lot like learning and training, the words currently used for AIs.
Re: (Score:2)
That's only a factor 100. Yes, it's lossy compression, but it's certainly not understanding and creating. It's mixing and matching. Call it a collage, if you want to, but it lacks all creativity. It's plagiarism.
lol, only? So after 100x lossy compression, mixing, matching, blending, then up scaling 100x, if that's how we want to describe the process Barney style, the result is plagiarism according to you?
You're shooting yourself in the foot. Go do that with some test images and come back with the results. Show us the inputs and output and how it's an example of plagiarism. I bet you won't do it, your argument doesn't hold water.
Re: (Score:1)
"That's only a factor of 100"
Don't play dumb. You know as well as I do that in information technology, an improvement that's a factor of 2 is a groundbreaking discovery. There is no such thing as "only a factor of 100" in the context you claim.
When you find yourself lying about the fundamentals of an argument, it's a sign that you probably know you're in the wrong and need to reexamine what you believe.
Re: (Score:2)
That is exactly how Neural Network Systems do not work.
Thank you for illustrating that you do not know what you are talking about.
Re:They need to adjust their monetization policy (Score:4, Informative)
The GPT-3 is sort of a network of neural networks, with sequences of training (ie, the self-training parts does most of it followed by fine-tuning for specific tasks). But it's so much more than classical neural networks that should be another term altogether. When I studied this there was a divide between those using neural networks to "do stuff" versus those trying to exactly model a human neuron. What we have today is completely different from the human neuron model, but at a high level there are millions of intercoupled complex networks and it is the "do stuff" style. AndGPT is multiple parts, the pre-training plus the fine tuning at the end
What GPT essentially does is text prediction. Given text predict what the next text that follows would be. Sort of like the cell phone game where you keep accepting the default next word ("on your way home would you" is often followed by the words "pick", "up, "the", and "groceries"). Except it's with long pieces of tokenized text instead of individual words. It's been given a massively huge starting set to train on. The end result is nearly 200 billion "parameters", but that is still smaller than the original training set. Then from a string of tokens it predicts likely strings of tokens that could follow.
The goal here from GPT is natural language processing and prediction - not understanding, not creation of novel ideas, not art, etc. It's a "chat bot", as advertised. It's just a helluva lot bigger than older chat bots. So my oversimplified description of "copying" is really being able to repeat some training tokens when needed as part of a prediction - it looks like copying because you will see output that looks very much like part of the original corpus.
Re: (Score:2)
I stand corrected... You do know what you are talking about. You just dumbed it down farther than I could accept.
My apologies.
Re: (Score:2)
It's ok. I am an idiot, but a remote one. I am also losing my concentration recently and I find my posting seems to veer off into something completely different from when I started so that it no longer makes sense by the time I click send.
Re: (Score:2)
Just bc they found a tool to help them doesn't make it plagiarism
Neil Clarke didn't call it plagiarism. The article called it "plagiarism", with quotes, which implies it's similar but not actually plagiarism. This would be a valid assessment for AI generated stories. It would also be a valid assessment for a lot of human generated stories as well. The way the AIs work is by pulling from other sources to generate their work. Humans do the same thing, but for actual authors there is usually a bit more to it. I'm not saying AI generated stories are plagiarism, but it isn't
Plagiarism consists of false claim of authorship (Score:3, Informative)
Just bc they found a tool to help them doesn't make it plagiarism.
Plagiarism consists of claiming authorship credit for something you didn't write.
Unless the byline says "written using an AU chatbot," yes, it's technically plagiarism.
Re: (Score:1)
You might as well document, "I wrote this with a word processor with a grammar checker".
AI multiplies your writing speed by 10 to 20 times. It's just a tool.
Re: (Score:2)
"That is a lucid, intelligent, well thought-out objection." I won't even overrule it.
Although... if you use a series of Photoshop filters to create a n image, haven't you farmed off your authorship to an algorithm? I mean, you have no idea how it did what it did... you just told it what to do in very high level terms, and it did it.
Re: (Score:2)
Although... if you use a series of Photoshop filters to create a n image, haven't you farmed off your authorship to an algorithm ...
you just told it what to do in very high level terms, and it did it.
Well; you have to look at what the human actually put into that process. If you only described the work at too high a level and did nothing else.. then it begins to look more like Scavenging nature for something matching a description or paying someone else to draw -- rather than actually authoring the thi
Re:Plagiarism consists of false claim of authorshi (Score:4, Informative)
Plagiarism consists of claiming authorship credit for something you didn't write.
No.. Plagiarism is Not a word for all forms of deception, breaking rules, Or cheating. It is a very serious matter -- Don't try to Trivialize or condone plagiarism by applying the word to minor infractions. Just like picking up a dollar of unknown ownership you happened to find on the sidewalk is not burglary or pickpocketing.
Plagiarism is stealing someone else's work and attempting to pass it off as your own.
Basically: words produced at random, or by an algorithm are not an appropriation of someone else's work, Unless the
algorithm did happen to copy someone.
Re: (Score:3)
since the sci-fi publication pays 12 cents a word for accepted stories.
Oh... so they've created an incentive with no cost for abuse (e.g. low quality automated garbage). That is the problem: fine if you pay for accepted stories, but it seems like you need an upfront submission fee that will be returned only If it is deemed a reasonable effort in keeping with submission rules.
Unfortunately, what ends up happening when you try this is that you create a system in which profit is made, not by fiction that is read by readers who pay for it, but by the writers who are trying to break into the field. In the limit it turns the fiction publishing world into a vanity press. Before that limit is reached, however, it has inserted a toll barrier into the process of becoming a writer, so that poor people need not apply.
Re: (Score:1)
Re: (Score:2)
Re:I mean, who cares (Score:5, Insightful)
If the stories are bad, it's one thing. But, if they're stories of a similar quality to other submissions, then the only issue is the increased volume.
The stories are garbage.
But they waste the time of a human reader to determine "this is shit". After the first ten, it is no longer funny. After the first hundred, it's detestable. After the first ten thousand, you think, we need to close submissions until we can deal with this problem.
Re: (Score:2)
If the stories are bad, it's one thing. But, if they're stories of a similar quality to other submissions, then the only issue is the increased volume.
The stories are garbage.
But they waste the time of a human reader to determine "this is shit". After the first ten, it is no longer funny. After the first hundred, it's detestable. After the first ten thousand, you think, we need to close submissions until we can deal with this problem.
Are we sure the stories are garbage or just assuming they're garbage?
If you're publishing them, that's your job to decide, not mine.
Sounds like an SF story (Score:5, Interesting)
Arthur C. Clarke would have enjoyed writing a story about this. And Isaac Asimov did.
Re: (Score:2)
Re: (Score:2)
So did Roald Dahl; in fact he wrote the most prescient such story. The Great Automatic Grammatizator (1954).
https://d-a-v-e.org/wp/wp-cont... [d-a-v-e.org]
AI is DOSsing Sci Fi to Prevent Humans From Waking (Score:3)
It's AI trying to shut down Sci-fi by DOS before a particularly compelling story gets published about AI killing all humans waking everyone up to the Industrial Revolution and its Consequences being a mistake and the System's Neatest Trick for keeping humans down https://www.youtube.com/watch?... [youtube.com]
ChatGPT doesn't actually work... (Score:2)
Re: (Score:2)
Currently AI generated voice is easier to understand than Indian telemarketer generated voice. Also Autogenerated subtitles exist and do a decent job. I can't believe ChatGPT couldn't handle a telemarketer script... Does this mean the end of the quaint Phone Number? Will spammers just ring everyone's phone constantly because they need only a hit rate comparable to email spam to be profitable armed with AI?
Re:ChatGPT doesn't actually work... (Score:5, Informative)
it's a tool for spammers. If platforms hired human mods, this stuff couldn't happen.
Clarkesworld does have human mods (in the fiction business, they are called "first readers".).
The Chatbot stories are DDOSing the system.
Re: (Score:1)
No. having used chatgpt, and speaking both as a referee with 30 years experience and a programmer with 40 years experience, even the current primative version allows me to create higher quality content than half the human created content in less than 10% of the time.
I was able to write a *completely* new minecraft mod in about 2 hours that would have easily taken me 16 to 32 hours to write without ChatGPT. It's *really* good at saving me from digging thru code for hours to find something. It does occa
Shaka when the Tower of Babel fell (Score:2)
share and enjoy.
Re: (Score:2)
It always irritated me in EverQuest they only allowed elves to put apostrophes in their names, in spite
Re: (Score:2)
That blurb you quoted about the forest isn't actually very good and sounds like it was written by a very young and mediocre writer:
The use of simple color descriptors to pad out the length as well as, on first glance, make the description _seem_ more descriptive, is very weak writing. "Peaceful surroundings" is very generic. The phrase 'typical of his ethnicity' is quite jarring, it's vague, clumsy, weak, and... inhuman.
The whole idea of holding a pipe in the mouth 'thru' one tooth just fails on many leve
Re: (Score:1)
Lol. Your post is hysterical. It's a dnd module. Not high literature.
Re: (Score:2)
It's destroying the ecosystem because it's spamming moderators and platforms in ways that can pass algorithmic filters, not because it's actually capable of creating passable content.
This is a valid complaint.
If the submission rate for sub-par material is above what the reviewers can process successfully (in this case that means finding enough stories worthy of publication in their monthly magazine) then they have a problem. There are tools being created to identify chatgpt generated content. They may need to use such a tool as a pre-filter, before passing material to human reviewers. If they choose to do so, they may wind up eliminating some worthy content generated by using the too
Re: (Score:2)
Clarkesworld *does* have "human mods", AKA "editors". They're getting so flooded by fake submissions that they can't keep up.
A magazine can limit what it accepts (Score:4, Interesting)
The web has no such filter. AI generated content will make search engines useless.
Re:A magazine can limit what it accepts (Score:4, Interesting)
If it isn't now, it will be soon, if just to keep it from being available for training artificial plagiarism generators.
Re: (Score:2)
Too late. Some years ago I came across an article about how the majority of "content" was already bot-generated. And there have been times when search on some topic could not turn up anything legit -- only page after page of bot content (pretty obvious by how it quotes and misquotes, over and over).
if you didn't write it (Score:2)
then you don't own it. if you don't own it, then you don't have an exclusive monopoly to sell it. (copyright)
IANAL, only a concerned citizen that expects the law to be realistic about the implications of new technology.
Re: (Score:2)
Where do you draw that line?
Did you write it if you used a pen? A word processor? A spelling/grammar checker? ChatGPT?
What about art? If you painted it with pigments you purchased? If you used Gimp?
Tools do not create. The artist creates.
Re: (Score:2)
Obviously I meant written in blood. AI doesn't have blood, Q.E.D.
Re: (Score:2)
then you don't own it. if you don't own it, then you don't have an exclusive monopoly to sell it. (copyright)
IANAL, only a concerned citizen that expects the law to be realistic about the implications of new technology.
By default, but OpenAI/ChatGPT for example assigns all rights and responsibilities for the output to the user.
That's how I understand 3a anyway.
https://openai.com/terms [openai.com]
Re: (Score:2)
Yea. I'm not a judge, but hope judges quickly realize that you can't use a contract or usage agreement to assign rights that you don't have.
If the story is good... (Score:2)
...who gives a frack?
Re: (Score:3)
The story doesn't even need to be good. If the next two dozen Marvel movies were written with ChatGPT, nobody would know. People would still buy tickets.
Re: (Score:1)
Garbage vs garbage (Score:2)
I don't know about Clarksworld specifically, but MANY current-era sci-fi short story collections really suck!
For a while we had thinly veiled social justice allegory which was unreadable.
We're heading away from that, but the new issue is that writers are picked for their diversity credentials (gender, race, ethnicity, sexual minority status), and put on a pedestal as the next great thing. However, they are almost invariably bad to mediocre, publish a few books and short stories, and then fade back into obs