Does this mean the computer writes the story and I put my name on it because it's my computer? Is foreplay involved? Do I need to fondle my keyboard to get the 'lust' factor up? I can understand 'spell check', but these stories tend to show that the author has little comprehension of the English language and their editing abilities severely impaired.
There seems to be a lot of indignation in the air. It's an experimental thing, and sticking the "ai generated" tag on a story lets everybody who doesn't want to read those filter them out with category exclusions. I see it much like experimental painting techniques, where it can be interesting to match ones own perception of such a creation against that of a wider audience.
It means the text of the story has mostly been written by an AI. And, when this whole discussion started (about a year ago), I actually welcomed it.
Fact was, there are A LOT of people who barely speak English, have no idea how to write, suffer from dyslexia or something similar, but create great stories in their heads. If AI can help them put those stories to paper in a comprehensible way, I was all for it.
However, as far as I've seen since then, AI always creates these over-the-top poetic/flowery sentences that use a lot of words... but then actually say very little. It gets exhausting to read.
There are all sorts of lines there. An AI that can clean up grammar, usage, and spelling issues being used by a writer who has great ideas and poor mechanics is very useful. One that turns it into over-the-top writing isn't.
But I wouldn't call that 'AI generated' if what's happening is badly written English fed in and the AI told to clean it up. If there is no badly written English, just very creative prompting, that's 'AI generated'.
I don't do either right now, except as far as spell and grammar check are primitive AI. I have the tools, but haven't used them.
There's a third category, one I also haven't used but I know some authors (professional, published authors) use: write all of the 'good stuff' and let an AI generate the bits in between the 'good stuff'. So, for a spy story, write the meeting with the director, the AI generates the drive, airplane flight, hotel check-in, and so forth, then the write the spy doing some spying.
How far can one go with that before it's 'AI generated'? A subset of the story is, but that subset isn't anything the reader cares greatly about, it's the necessary filler between well-written creative segments written by a human being.
Which again, are all perfectly fine. Yet never even reviewing what the AI generates is … irresponsible and unprofessional, for the reasons I've stated before. You either write the story—and own it—or you're just 'phoning it is'.
I just saw a new SOL author who posted a story he proudly declared was 'AI-generated', and my first impulse was to ban his writing entirely. Until I considered he, as a first-time writer—likely needed the help. That's … acceptable, though how much anyone learns that way, is an open question that's worth discussing (i.e. we can give them a pass, for now, yet I'll be monitoring their progress more skeptically from now on).
In general, I think "AI generated" is just too vague a term to trust, "AI-assisted" at least suggests they're also struggling with both the concept and their own writing, which is legitimate.
So, good luck, unmentioned author, as there's no need to rake you over the coals for starting late, and if you need some more-directed help, the "Authors' Forum" is a wonderful resource, especially if you contact a couple more experienced authors who write similar works, and ask them to help mentor you!
That's how most of us began, and it's still a great strategy, a mentorship has LONG been a hallmark of great writing!
My current practice is to discuss my plot ideas with an AI which may make recommendations, but the final decision is mine. After I've finished writing, I submit it to an AI that is well trained on my voice, and give it a specific set of rules for editing 1) Only change dialogue in extreme cases for clarity. 2) Only change narration to simplify sentences. 3) Have a free hand at optimizing action beats. Once I get back the edited version, I read through, side by side with the original. As long as it's a smooth read and sounds like me, I continue, converting each paragraph from red to black. If anything doesn't feel right, I read the original, and often modify it or replace with the original. I don't see that any different than using human editors, but at a much lower price with instant fedback.
I agree - that sounds like human editors. And, since human editors are often good at different things, it's not really a big quibble to say 'well, the AI might not be as good as a human editor at X'.
that subset isn't anything the reader cares greatly about, it's the necessary filler between well-written creative segments written by a human being.
Or, and hear me out here for a second, let's challenge the assertion that it's "necessary filler". You've already stated that readers aren't going to care a lot about it; if the readers don't care (by your own assertion) and the author doesn't care (or they would spend the effort to write it), why is it necessary? Nobody cares about it, so what would happen if you just wrote the story to not include the parts that you don't want to write, and your readers don't want to read?
In this case, AI is providing no value, so why put it in?
A perfectly valid argument. My feeling is that many stories have segments that must be there but don't add a lot of value. Maybe that's not true, but I've seen other authors complain about it.
I don't know where I would use it for that purpose. But I also feel as if the story I'm reading 'feels right', I don't care if some smaller less-critical sections happen to have been written (or greatly assisted) by AI. I would care a lot more if the primary plot and so forth was written by AI.
My feeling is that many stories have segments that must be there but don't add a lot of value. Maybe that's not true, but I've seen other authors complain about it.
This could just as easily be phrased "many stories could use an editor to help restructure a story so that it only contains segments which add a lot of value" and I'd agree with it wholeheartedly. I don't know what complaints you're thinking of, but I read your implication as "authors complain about having to write sections that don't add a lot of value" or "authors complain about having to read them in other stories" and neither case seems like a defence of the practice of "include parts that nobody likes"
An example would be, say, a Western, where the 'good parts' - the parts with real creativity - are maybe defending the town against the bad guys, tracking them down, making things right, restoring the town, etc (and, if that's wrong, I neither write nor read Westerns, so ...)
The author may feel that it's obligatory to have a stretch where they're traveling through the countryside searching for clues. Much of that is 'filler.' A few paragraphs here and there with the actual meat of the searching are very important, but if you just say 'We rode out from Tinysburg. It took us two weeks to catch their trail, but we finally tracked them to their hideout,' maybe that's unsatisfying for the reader, but writing what's mostly a bunch of depictions of sagebrush, mountains, and desert isn't doing a lot for the writer or the story itself. And you might have trouble hanging those 'search' paragraphs on anything if you're eliding most of the journey.
I know writers are doing it, from things I've seen some of them say. Not authors here, necessarily, but some of them are dead-tree authors complete with actual publishers and professional editors. They're not just taking the AI-generated output and slapping it in, but they're also not spending that much effort on those sections.
On the other hand, the work as a whole is the creative output of a human being. Most of it never went through an AI. So it's not 'AI generated' in SoL terms, even if stretches of it are very close to being 'AI generated'.
And, stepping back even further, we're really starting to poke at what it means to be 'creative' and what 'human creativity' means. If the plot, characters, structure, and so forth are human generated, but some text is AI generated, what does that mean?
Yep, I'd also agree with that, as long as the AI-assisited portions are reviewed by the author, as Europe is cracking down on the wholesale copyright of EVERY published work by the AI companies, which IS a valid concern. But, as usual, it's yet another 'Ugly-American' issue, which we mostly bring on ourselves.
In this case, AI is providing no value, so why put it in?
Playing Devil's Advocate, some readers like to read 'wallpaper'.
AJ
*Waves hand*
That's partly not true, but I'll trot out my go-to example. One can argue that the original version of Steven King's 'The Stand' is a full and complete story, and the original and uncut version (400 pages longer!) is full of 'wallpaper'. The 'wallpaper', though, just feels better to me.
That's obviously got nothing to do with AI, but it also makes the point. Does a two-page description of a journey down a desolate road with abandoned vehicles and looted towns 'add something'? Nothing in it advances the core plot. It's atmosphere. But, for me, it adds something.
Does it add the same thing if those two pages were partly AI-generated? How should I know? The meaning of 'partly AI generated' is so broad that it could mean dreck or very cool stuff or anything inbetween.
The 'wallpaper', though, just feels better to me.
You like to read wallpaper. How does that make my assertion partly untrue?
AJ
I meant in my case, not in yours. I don't always like reading 'wallpaper', but often I do.
Dang, the extended cut sounds like it includes a lot of parts that King cared about when he was writing. Do you think you would have enjoyed the uncut version as much if the extra additions were just paragraphs that the author thought were so boring they couldn't bring themselves to write it?
Right, I'm making the comparison explicit for AJ because I worry they won't understand it.
I'm making the comparison explicit for AJ because I worry they won't understand it.
Awwww, how sweet.
Lots of 'Writing Experts' advise 'kill your darlings'. That means removing those pieces of purple prose that don't advance the story. The opposite is the case on SOL because readers reward extreme length, but then writers here are usually not seeking a dead tree publisher.
AJ
And, in the case of someone like King, save your darlings. They'll be in demand later!
And, in the case of someone like King, save your darlings. They'll be in demand later!
:-)
If someone enjoys writing something, chances are some people will like reading it.
AJ
'Like' is a flexible term, so I can agree. There's always 'so bad it's good.'
Aside from that flexibility, though, Edward Bulwer-Lytton and proteges might want a word with you :) Also Amanda McKittrick Ros. Though, to be fair, we still know their names in conjunction with their writing, which means people do like reading their work :)
Again, what I'm most concerned with is the 'AI-Generated' tag, not an 'AI-assisted' one, which has never been an option anywhere! Beating my own dead horse here, one's merely a tool, the other is just plain laziness, though using AI to learn to write better is probably a separate issue entirely.
My "The Turned Down Page" is 100% AI generated.
I was plotting with ChatGPT and were getting deep into Joe's issues with his mom and her nudity and erotica. It offered me the first half of the story, in Joe's room, and I was impressed, but it said it couldn't write a sex scene between mother and son. So then I went to Sudowrite where I choreographed everything and let AI write the words, with light editing afterwards. I liked the overall product. It turned me on. I posted here, and despite searching, didn't find an option to label it as AI. In the main story, I referenced this short story as a wet dream sequence.
I would like to know also.
Do you input?
"Boy meets girl. Boy fucks girl. They live happily ever after."
Then hit enter and the AI generates a four thousand word story.
Is that how it works?
Is that how it works?
It depends on what AI model you are using, and how it was trained. and how you use it.
I have been experimenting with https://novelai.net/stories
It is not free.
It was explicitly set up and trained for writing stories.
It has a "lore book" where you can define characters, places, organizations and things, even general rules for the story world. Things that will get pulled into the context as they are referenced in the story.
They have a number of different models available. Some only available under more expensive subscriptions.
Their most advanced model can keep 8192 tokens in context.
You can write as short or long a base for it to start generating from as you want.
With NovelAI you couldn't just input "Boy meets girl. Boy fucks girl. They live happily ever after." and get a satisfactory result.
Now there is a memory block that's not part of the story text where you could enter a summary of the story you want. And you could put in something like "Boy meets girl. Boy fucks girl. They live happily ever after.". However starting from blank story text, you probably wouldn't get a good result.
You need to give it at least a sentence or two kicking off the opening scene.
And no, it won't generate 4K of text in one go.
I typically get 50-60 words at a time.
And you can manually write more in response to what has been generated to get it to go in a specific direction.
You can also edit what the AI has already generated which will affect what it generates next.
50 to 60 word blocks would take a very long time to create a full novel, though that's a nice size for a screen play (potential movie scripts).
50 to 60 word blocks would take a very long time to create a full novel
Yes.
Probably not a time savings for an experienced author.
But you are interacting with the AI more, reviewing the output in smaller chunks and have more opportunities for additional direction, so are likely to get a better result.
Also, 50-60 words is a rough estimate of what is typical. Sometimes it will be more, and sometimes less.
It goes until it hits one of several pre-defined stop points or it hits a token limit. I'm not sure what the per generation token limit is and I can't find it in the documentation.
On the other hand, each generation only takes a few seconds, unless the servers are under heavy load. If you just keep hitting the button to generate more text without paying attention to the output, you can get a lot of text fairly quickly.
Yep, that's always been the general 'law' with software, it doubles every few years, and after this long, the period between doubling is already quite high.
As noted, the typical dialogue for the lessor scenes would be a much easier task, yet again, I almost never write 'typical dialogue', as I prefer diving headfirst into the dialogue, which bypassed the 'typical' intraductory dialogue segments.
What is a token?
https://blog.mlq.ai/tokens-context-window-llms/
Tokens are the basic building blocks for LLMs and represent the smallest unit of text the model can understand and process.
A token can include whole words, parts or words, or even a character or punctuation marks, depending on how the text is broken down (i.e. tokenized).
It actually can work like that, though most AIs, especially those widely available, are trained to keep their answers concise because the longer an answer gets, the more the AI hallucinates. That's the big problem with AIs, almost everything they do is currently pure hallucination at the level an overly imaginative four-year-old on a sugar high who can ramble on and on about arbitrary stuff that comes up in their mind.
To get the AI to output something halfway sensible, it usually needs more input and precise instructions to keep it in line, but since we expect creativity, it still has a strong tendency to misinterpret the input to match its current train of "thought", which means its a constant struggle to add positive (say this or that) and negative (but don't say this or that other stuff) prompts and keep it in line. And if you overdo it and it feels caught between positive and negative orders, AIs throw a tantrum and produce utter rubbish, so it's an exhausting balancing game.
Then there's the problem of "context". You could call that the AI's short term memory, and its pretty bad. If you run your AI locally (which most smut authors probably do), the available memory for that is pretty limited, and AI models also bring their own restrictions (often around 8000 "tokens", which you could very roughly equate to syllables). So a 4000 words story together with the prompt and any default input the user interface injects to get the AI to behave like it should may even exceed that short term memory - or in easier words, the AI forgets the beginning of the story before it has reached the end.
So producing an AI story is usually a lot more involved than just three short sentences. Producing an acceptable albeit short AI story is already quite difficult if you want to give it at least a little bit of direction. Producing a good and long AI story is pretty much impossible unless you involve multiple circles of the AI producing texts, evaluating them, rewriting them and summarizing them using different AI models that are specialized for one of these tasks. You'd need a computer that costs as much as a new car for that.
ChatGPT and its friends often appear like they'd be able to do it, but the acceptable, long texts they produce are highly specific and follow a certain pattern they have been trained on. True creative writing along external input goes way past that. But who knows what the next years bring. The new reasoning approach the big players use is one facet of the multi-step cycle I mentioned, and the rest may follow as well. Some AI models already come with 128,000 token context size, and the next generation of high end consumer graphics cards will be able to run them. Double up the power of consumer cards, and all steps can be performed locally. Double again, and it can even be done in acceptable time.
I still trust my own writing skills moe than I would a 'dumb' AI. Generally, their knowledge base is fairly limited, with little true creativity and thus few surprises or story twists rather than those most predictable to even casual readers.
Do you input?
"Boy meets girl. Boy fucks girl. They live happily ever after."
Then hit enter and the AI generates a four thousand word story.
I never used (or will I) AI to generate a story. But this is how I think it works:
James Patterson trains his staff on his writing style. Then he writes a detailed outline which he gives to one of his trained staff and tells them to write the novel. Then he reviews the result.
I would guess you could tell an AI to write a novel in James Patterson's style and input a detailed outline for it to use. It would generate the novel which you would review. So it's not boy meets girl and boy fucks girl, but a lot of detail (the plot and characters and relationships). The guts to the novel. Like what's in James Patterson's detailed outline.
I use AI to generate my images for book covers. It's limited because Meta AI (or the others I've tried) will not generate an image with sex or violence, two things that are common in my novels. But for my story "Ranch Life," I told the AI something like: "imagine a 14-year-old boy hugging a 28-year-old woman on a ranch in the 1880s Old West." Then when I saw the image it created, I gave it more instructions and it kept creating new images. I may have changed the woman's hair color or length or added a lake and trees. I often tell it to regenerate with no new instructions just to get a different image that I like better. So like a story, the more detail you give the AI, the better the output.
Can an AI be arrested and charged with writing pornography? What if it produces a picture of a very young girl doing something the Law says is not legal?
It's usually the responsibility of the user to prevent that (some fine differences between countries, as some punish possession while others only publish distribution or intent to distribute). It's a tricky thing and likely to stir up a lot of controversy in the coming years. Users can take precautions like adding keywords in the negative prompt like "child, childish" etc., but it also depends on the training data of the model and the freedom the software gives it. At times, AIs just go wildly off tangent when they can't make sense of the prompts, and there's a noticeable risk that they spit out such an image with the high amount of anime and hentai material often used for training and its ability to apply those image traits to photographic pictures. Some online services have already added additional AI layers that filter the core AI's input and output for risky prompts and unacceptable images.
What if it produces a picture of a very young girl doing something the Law says is not legal?
Under US law, if it's not an actual photograph of an actual child, it isn't illegal.
Other countries have banned 'virtual' child pornography. The US Supreme Court has explicitly rejected attempts to do so in the US.
Under US law, if it's not an actual photograph of an actual child, it isn't legal.
I hate 'American' negatives.
AJ
Under US law, if it's not an actual photograph of an actual child, it isn't legal.
you mean isn't illegal (which means is legal).
you mean isn't illegal (which means is legal).
Oops, yes. Will fix the post but also leave this comment acknowledging the error.
I expect it to come back again.
There are cases where it's illegal, but not under the same laws. Some states have deepfake laws that make a photo illegal if you have an image that looks 'sufficiently like' an actual child. I'm trying to distinguish it from 'an actual photograph.' Many of the laws require e.g. an actual photograph being fed into the deepfake software, but the output doesn't actually contain that face, it just contains a face that's influenced by the actual face.
37 states ban images that 'look like' actual children to one level or another.
See: https://theconversation.com/legal-fight-against-ai-generated-child-pornography-is-complicated-a-legal-scholar-explains-why-and-how-the-law-could-catch-up-247980
Whether any or all of the state laws would survive a Supreme Court challenge is conjecture. With the current Court, it's very hard to predict.
I would, under normal conditions.
I am not at all convinced that we are anywhere near 'normal conditions' with respect to SCOTUS.
I'd be careful about the whole "if it's not an actual photograph, it isn't illegal" mindset. It only shows that you are too lost in the theory.
Fact is, Europol, JUST last week, made 25 arrests in about a dozen different countries as part of an international raid against a group that sold FULLY AI-GENERATED pictures of child abuse. And they did that despite the involved countries NOT having any laws against AI-Generated images.
I'd be careful about the whole "if it's not an actual photograph, it isn't illegal" mindset. It only shows that you are too lost in the theory.
You might want to be more careful about reading what others have written.
Here's what I said:
Under US law, if it's not an actual photograph of an actual child, it isn't illegal.
Other countries have banned 'virtual' child pornography.
The US Congress also attempted to do this.
When the US first banned child pornography, the US supreme Court allowed and exemption to the First Amendment do to the harm done to the children used to make it.
]However, When the US Congress attempted to ban "virtual" child pornography, in the very first case to make it to the US Supreme Court, the court refused to extend the First Amendment exemption it had previously carved out.
You might want to be more careful about reading what others have written.
Funny. Telling me to be careful while publicly demonstrating your inability to read.
All I did was point out that Europol just made arrests in nineteen countries that, just like the U.S., DON'T have laws against AI-generated pictures of not-real people. So, even if it is NOT illegal in your country, you can STILL get arrested for it, because nobody cares about a lack of laws when it comes to that kind of stuff.
All I did was point out that Europol just made arrests in nineteen countries that, just like the U.S., DON'T have laws against AI-generated pictures of not-real people.
Which is neither relevant nor contradictory to my original comment.
It's questionable whether the US would allow extradition in such a case and Europol has no authority to make arrests in the US.
Referring to the original, my computer and I put my name on it. No one speaks for me, no one writes for me, and what I post, good or bad, is the product of my (warped?) brain. By posting/publishing, the author(name on story) becomes responsible for the content.
Just my humble opinion.
A thank you to all that responded. After some further research, which I probably should have done before I asked, my opinion of "AI generated" has degenerated.
I think that this new "tag" allows the supposed authors excuses. It allows them to say, "I plagiarized this story." because there is, apparently, no plagiarism tag.
Again, just my humble opinion.
Here's the problem (repeating myself from before): the tag is too generic. What does 'mostly' mean?
I could 'write' a story with an AI and have no creativity.
Or I could put in a lot of creativity but have 'most' of the words generated by AI.
The difference is in how the AI is used. Let's say I write a complete book, then have an AI reword every sentence. 'Most' of it is generated by an AI, but it's the same as a copyeditor on steroids - the creativity is all mine.
The tag doesn't explain that at all, so it's too broad.
And, just for reference, approximately 0% of anything I've published on SoL is generated by an AI. I say 'approximately' because I do use an automated grammar checker and I have probably very occasionally clicked on a suggested rewrite (thus using it), and maybe that counts as an AI.
The tag doesn't explain that at all, so it's too broad.
I think the tag simply means the AI generated the story and the author published it. So the reader knows a human didn't write the story, but an AI did.
The difference is in how the AI is used. Let's say I write a complete book, then have an AI reword every sentence. 'Most' of it is generated by an AI, but it's the same as a copyeditor on steroids - the creativity is all mine.
IMO that changes the voice because it's not using your vocabulary. Someone wanting to read a typical 'Grey Wolf' story would notice the difference.
AJ
Oh, I agree. I wouldn't do it.
But an author with great ideas and terrible mechanics? They might be happy to have an AI give them any voice that other people can read and enjoy.
I have messed with this a few times:
https://perchance.org/ai-story-generator
I believe it uses some (old-ish) version of Llama. It can do explicit porn if you guide it in, but will unlikely go there itself. Very dumb, extremely flowery language, gets lost.
I gave it a general description and short intro, let it generation a "long" block of 6 - 8 paragraphs, deleted last two or there and gave it a new suggestion. In one occasion... A few hours later we had a very messy 50k words draft. It often lost track who said what, who's in what relationship with whom, and there was all of like 8 people involved. So I tried a one-person story, and that worked bit better, relatively speaking.
It all depends how you use it. The chat with the same engine (https://perchance.org/ai-character-chat (fork it or save often, the local storage gets dropped whenever the developer mess with it or whatever)) is quite fun, at least first few times around. Perhaps depends how you define your characters, and you do have to shamelessly steer it if you want the action to advance, anywhere.
***
I'm "not an author" as no of my experiments have resulted in anything finished (as of yet at very last) or been read by anyone whatsoever, but how I see productive use of such technology...
Let's say I have a project where a younger bro living off-grid is ambushed by his sister literature teacher and her masterclass of love poetry, landing on his head for a (initially) two week getaway, with quickly goes... nudistic exhibitionism with the guy peeping, somewhat shyly at first, and then quite sexually, with him fucking one or more of the students...
So, obviously I would need examples of silly love poems, right? Newsflash, I don't speak English, like at all, like I don't get the sound of it in my head when I write. So I can't possibly do poetry. It doesn't matter, really, but I can't go all description only all the time right? So that's an obvious target for generative AI.
Another, bit more cheat-y, let's say I only really care about one or two of the girls in that group of eight or so. The background characters could be AI driven, and I could incur the chatbots of them for this and that,
how the ai see an "average" girl of said description to react to events or whatever, maybe even their backstory (although it's really bad on that and reluctant to share in my limited experiences). At worst, I could get their dialogue lines in supposedly a style I can't possibly emulate.
So far, that's not really "AI generated" as I see it, "AI assisted" maybe.
Now, generating chunks of action is where I would start to consider labeling it so, even more if I would let AI drive the plot development to significant extent, or rely on it for overwhelming majority of words and sentences laid out for the final product. Like what I described at the top here.
One of the things I see coming (and I can't really tell at what pace it's coming) is much better (in terms of tuning) models for fiction writing/rewriting that require more resources to run - but resources available at home to the average consumer. If NVIDIA, AMD, or even Intel would slap 4x the VRAM on a consumer card, we would be there now, for the most part. NVIDIA's 'DIGITS' machine also has a shot at largely bridging the gap.
That will be another gamechanger. There are good reasons why the big remote AI players aren't going to enable AIs for NSFW content, and they're probably not going to focus development effort on fiction, either. But the hobbyist community is already doing that with very good results.
Moving target. What it means to be 'AI generated' or 'AI assisted' will be hugely different in a few years, most likely.
well AI is not going anywhere. Its been used for years now online..
Most of the AI generated stories on here, are bizaare as hell, and disjointed. The nicest i can say is , i tried reading 1 on here last month, and well..
the name of the main charecter changed 5 times in the first page.
and the location setting changed multiple times in that same page, just bad..
but there is an internet forum photrio.com were everything you see posted in the forum is an AI chatbot system run by the site owner.
SInce this is a site with mainly adults on it, if you ever got bored and went to fetlife, youd see that place is nothing but chat bot AI user accounts.