Cousin topic = incest?!?

[KeithD]

That I/T is a trump category here has been the general understanding since, like, forever. It's not exactly a revelation.

 

[ElectricBlue]

From the bolded part above, I take it your story was rejected, or placed in a category not requested the first time? Or did you explain it in the comments section with the original submission?

If it were #1 above, it supports my statement, that I&T trumps all. I never discounted the fact that Laurel wouldn't move it to where you asked it to be. She runs the place and has that power. She can pretty much do what ever she pleases.

Comshaw

No, it wasn't rejected - it went through first time and was categorised as I'd requested. I explained my reasoning in a Note to the Editor, and Laurel obviously agreed - there was no to and fro, and no unusual delay in publication, either.

I&T is nearly always a trump, but it's not an absolute.

 

[Comshaw]

No, it wasn't rejected - it went through first time and was categorised as I'd requested. I explained my reasoning in a Note to the Editor, and Laurel obviously agreed - there was no to and fro, and no unusual delay in publication, either.

I&T is nearly always a trump, but it's not an absolute.

You are correct on that point. There are rarely absolutes, especially when dealing with human judgement.

Coimshaw

 

[KeithD]

Even in the spelling of our own names, apparently.

 

[ElectricBlue]

You are correct on that point. There are rarely absolutes, especially when dealing with human judgement.

I was correct on my original point, too .

 

[Comshaw]

I was correct on my original point, too .

Sorry, but I must have missed something. Your original point was, what? As far as I can tell, it was about absolutes. Would you enlighten me on what I missed?

Comshaw

 

[ElectricBlue]

Sorry, but I must have missed something. Your original point was, what? As far as I can tell, it was about absolutes. Would you enlighten me on what I missed?

Comshaw

My original post made the observation that not all stories with incest themes get trumped into the I&T category (which is what you had stated), as evidenced by my story that went where I asked for it to go, first submission, no debate, no rejection.

I then observed, later, that "I&T is nearly always a trump, but it's not an absolute," to which you replied etc...

 

[Comshaw]

My original post made the observation that not all stories with incest themes get trumped into the I&T category (which is what you had stated), as evidenced by my story that went where I asked for it to go, first submission, no debate, no rejection.

I then observed, later, that "I&T is nearly always a trump, but it's not an absolute," to which you replied etc...

Aw, I see the reason for my confusion.

If you would be so kind as to reread your above quoted statement, you'll find you did not make two points, only one. The two bolded parts are essentially the same. As I said, you were correct in that singular observation.

Comshaw

 

[gordo12]

Reading through this thread has me reconsidering the category for my V-day story. Although originally written as a romance, it features two people who could be related as in identical twins (but born to different identical twin mothers, but fathered by identical twin fathers) or first cousins, or a random fluke of matching DNA. (one in six billion). It's a stretch but a fun romp.

Although I trickle breadcrumbs throughout the story suggesting that maybe they are related...at the end, I dodge away from answering the question. It's left up in the air. Sort of no one wants to air the dirty laundry.

Reading through, I'm left with the thought that romance readers won't like it for that reason, and for incest readers, it doesn't go far enough. The couple do get married because no one, including them, knows about it.

Thoughts?

 

[Dingo_Triplesix]

Exactly my thinking. If you just write for your own pleasure, you can do whatever you like. But if you want to fulfill the expectations of the readers, you have to take into account the preforming of the category system.

 

[Bramblethorn]

AI can do a lot of things, including writing stories. Erotica would be quite amenable to AI, since there are certain things readers expect and those wouldn't be that hard for AI software to identify from higher scoring stories in each category and to reproduce.

This has been discussed repeatedly on this forum. Tools like GPT-3 are impressive toys, great at style mimicry, but they have no understanding of internal logic. A short excerpt from an AI may be indistinguishable from the work of a human, but anything past a few paragraphs and the lack of understanding starts to become obvious. AIs may be able to help write stories, but I've yet to see anything that could write something suitable for Literotica without a great deal of human hand-holding.

If there is something, I'd be interested to see it. I wouldn't be at all surprised to see that technology develop, but I don't think we're there yet.

 

[NC_Coastal]

Writers write. Writers also like to argue with each other. These are absolute truths.
Oh, and yes, cousins screwing cousins is first-degree incest. At least on Lit.

 

[windar]

This has been discussed repeatedly on this forum. Tools like GPT-3 are impressive toys, great at style mimicry, but they have no understanding of internal logic. A short excerpt from an AI may be indistinguishable from the work of a human, but anything past a few paragraphs and the lack of understanding starts to become obvious. AIs may be able to help write stories, but I've yet to see anything that could write something suitable for Literotica without a great deal of human hand-holding.

If there is something, I'd be interested to see it. I wouldn't be at all surprised to see that technology develop, but I don't think we're there yet.

What is "suitable for Literotica"? You program in the elements that the average reader wants in a story in a given category. The programs already know the rules of grammar. The screening that stories receive here is not looking for depth of character or plot consistency; it's basically looking for formatting and not violating certain site rules and even this is done inconsistently.

So, my contention is that producing a story "suitable for Literotica' is very much within the capability of current AI. That doesn't mean they'd be great literature or even that you or I would like them, though I am quite confident that some readers would

 

[Bramblethorn]

What is "suitable for Literotica"? You program in the elements that the average reader wants in a story in a given category.

Let's unpack that a bit: what specifically does "programming in the elements" look like? How would you even define those elements, let alone tell your text generator to produce them?

For instance, let's take it as agreed that Loving Wives readers commonly like stories where cheaters get their come-uppance, Lesbian Sex readers like stories with female-female content but very little male-female, Romance readers like stories with happy endings, and so forth..

How would you go about telling GPT-3 (or some other text generator of your choice) to produce those different flavours of content for the different categories?

The screening that stories receive here is not looking for depth of character or plot consistency; it's basically looking for formatting and not violating certain site rules and even this is done inconsistently.

The moderation process isn't looking for those things, no. But we were talking about the readers. Literotica might not be the most demanding place out there where things like plotting and characterisation are concerned, but readers still want something that makes vaguely coherent sense, even if it's just something they can visualise while they stroke.

So, my contention is that producing a story "suitable for Literotica' is very much within the capability of current AI. That doesn't mean they'd be great literature or even that you or I would like them, though I am quite confident that some readers would

Okay. Show me a story of at least 3500 words (approximately one Literotica page), written by an AI tool with minimal human intervention, which you think is coherent enough to satisfy the average Literotica reader. Not asking for erotica here, just something at a similar level of story-telling.

People keep arguing that this is possible, and yet examples are distinctly lacking. Most of the examples I've seen have been short, no more than a few paragraphs, because the limitations of such methods become obvious over longer pieces. The rest mostly turn out to have heavy human intervention - "I generated twenty different paragraphs to go here, picked the one I liked best, and repeated until I got a whole story".

 

[TadOverdon]

This has been discussed repeatedly on this forum. Tools like GPT-3 are impressive toys, great at style mimicry, but they have no understanding of internal logic. A short excerpt from an AI may be indistinguishable from the work of a human, but anything past a few paragraphs and the lack of understanding starts to become obvious.

It does sound as if AI could be used to generate reader comments.

 

[windar]

Let's unpack that a bit: what specifically does "programming in the elements" look like? How would you even define those elements, let alone tell your text generator to produce them?

For instance, let's take it as agreed that Loving Wives readers commonly like stories where cheaters get their come-uppance, Lesbian Sex readers like stories with female-female content but very little male-female, Romance readers like stories with happy endings, and so forth..

How would you go about telling GPT-3 (or some other text generator of your choice) to produce those different flavours of content for the different categories?

You don't program AI with concepts. You program it with all of the stories in a given category (you could set some lower limit on scores if you like). It then tries to reproduce those stories. In a sense, this is similar to what we all as human (mostly) writers do. We read and assimilate some things from that and then put our own spin on it.

The moderation process isn't looking for those things, no. But we were talking about the readers. Literotica might not be the most demanding place out there where things like plotting and characterisation are concerned, but readers still want something that makes vaguely coherent sense, even if it's just something they can visualise while they stroke.

Okay. Show me a story of at least 3500 words (approximately one Literotica page), written by an AI tool with minimal human intervention, which you think is coherent enough to satisfy the average Literotica reader. Not asking for erotica here, just something at a similar level of story-telling.

People keep arguing that this is possible, and yet examples are distinctly lacking. Most of the examples I've seen have been short, no more than a few paragraphs, because the limitations of such methods become obvious over longer pieces. The rest mostly turn out to have heavy human intervention - "I generated twenty different paragraphs to go here, picked the one I liked best, and repeated until I got a whole story".

I have no doubt the first attempts would be clumsy. The same is true of many of us. We (hopefully) improve and so will the AI. The human guidance needed will diminish over time to very possibly minimal. So, you are probably correct that the program wouldn't produce much without some human intervention, especially at the beginning, but that isn't really the point.

Is anyone going to bother for stories on a free site? Probably not for other than an academic exercise. But think of the people who churn out 2 or 3 "shorts"/week and post on Amazon. Some claim they make a living, but it's a not a plush one. Now let AI take a good part of the work off the human, though not all. Now you could churn out a dozen stories/week or more. For a while that would work, until everybody does it and the market becomes totally staurated...

Now, will they be great literature? At present capabilities, probably not. But are Lit stories or Amazon shorts great literature? I think it would be feasible to train an AI program to turn out competently done erotica stories (other genres as well). And AI is getting better and better and humans aren't. Individually we can improve, but it's not clear that literature as a whole is better today than it was in Shakespeare's time or in classical times.

As for your challenge, if the programs weren't so expensive, I might take you up on it...

 

[8letters]

You don't program AI with concepts. You program it with all of the stories in a given category (you could set some lower limit on scores if you like). It then tries to reproduce those stories. In a sense, this is similar to what we all as human (mostly) writers do. We read and assimilate some things from that and then put our own spin on it.
:
I have no doubt the first attempts would be clumsy. The same is true of many of us. We (hopefully) improve and so will the AI. The human guidance needed will diminish over time to very possibly minimal. So, you are probably correct that the program wouldn't produce much without some human intervention, especially at the beginning, but that isn't really the point.

Is anyone going to bother for stories on a free site? Probably not for other than an academic exercise. But think of the people who churn out 2 or 3 "shorts"/week and post on Amazon. Some claim they make a living, but it's a not a plush one. Now let AI take a good part of the work off the human, though not all. Now you could churn out a dozen stories/week or more. For a while that would work, until everybody does it and the market becomes totally staurated...

Now, will they be great literature? At present capabilities, probably not. But are Lit stories or Amazon shorts great literature? I think it would be feasible to train an AI program to turn out competently done erotica stories (other genres as well). And AI is getting better and better and humans aren't. Individually we can improve, but it's not clear that literature as a whole is better today than it was in Shakespeare's time or in classical times.

As for your challenge, if the programs weren't so expensive, I might take you up on it...

I don't know much about AI, but from what I know, the idea of just feeding some stories into an AI program and expecting it to spit out coherent stories is a pipe dream.

Let's say that you wanted to write some software to write the very common I/T stroke story of a mom/sister/daughter/aunt/cousin/niece walks in on the son/brother/daughter/aunt/cousin/uncle masturbating, sees how big his cock is, and decides that she wants to be fucked by it. I don't think that'd be a huge project, but wouldn't be done with AI. You could have a web page for users to generate the story, they'd select the various options and the program would generate a short stroker that fits what they requested.

I'd think the first step to having AI write erotica would be for there to be an AI that would recognize the different parts of a sex scene - foreplay, oral and fucking. Then you could feed the AI oral portions of sex scenes to learn how to write similar portions.

 

[Bramblethorn]

You don't program AI with concepts. You program it with all of the stories in a given category (you could set some lower limit on scores if you like).

This isn't "programming it with the elements that the average reader wants", then. It's feeding it whole stories, and hoping that the black box will figure out how to extract those elements and recombine them into new stories.

Given that it costs something like 4 million dollars to train GPT-3, you really wouldn't want to retrain separately for every genre you want to write, and you might not have a large enough corpus to train it if you did. This might be solvable, though; given an opening paragraph or two, GPT seems capable of recognising genre.

It then tries to reproduce those stories. In a sense, this is similar to what we all as human (mostly) writers do. We read and assimilate some things from that and then put our own spin on it.

Similar in a very broad sense, but the details of what and how we assimilate are different in important ways.

Humans use words to convey meaning. When we read words, we build a mental model of the "reality" those words are describing, and often we remember that model long after we've forgotten the specific words. We can separate style from meaning, making it possible for us to do party tricks like "rewrite this passage in the style of Charles Dickens".

Some AIs attempt to model truths, others emulate style. But so far, I'm not aware of one that does a good job of integrating the two. Most of the text generators are sophisticated versions of "words are often used in these patterns".

I expect this problem will be solved, sooner or later, but I don't think we're there yet.

For an example of what I'm talking about with modelling truth, see GPT-3 generated samples here: https://www.gwern.net/GPT-3#harry-potter-and-the-methods-of-rationality

Excerpt 1:

The only magical creature which has eluded capture by the finest armies and best generals of all time. One of these creatures can still destroy a city. A hundred of them can destroy a country. The deadliest magical creature that exists—is a wizard or a witch who has decided they have nothing left to lose. A muggle with a gun can shoot down a protected wizard like a duck in a fairground. There is no defence. The only security against a wizard or witch who has decided they have nothing left to lose is to never let them lose in the first place."

“Never give your enemy a fair chance.” Professor Quirrell’s voice was now a low hiss, and even from his position near the back of the classroom, Harry could sense an undercurrent of rage in his words. “Because if you do give them a fair chance, they will take everything you love. If you give your enemy the choice to die for a cause, or to kill for a cause, they will always choose to kill. Never let your enemy have the chance to give everything they have for their dream, because they will take everything you have and then burn that too. The most powerful weapon in your arsenal will not be a jinx or a curse or a single spell. A wand can be broken. You can run out of magic. You will always have more strength and speed and cunning. The most powerful weapon in your arsenal will be the decision to fight, and the conviction to see it through. If you fight like you’re defending yourself against a mountain troll—if you fight like you’re defending your homes, your families, and your children, you will win no matter how powerful your enemy.”

First thing I'll say is that what GPT-3 does do is staggeringly impressive. It does a fantastic job of matching the way humans structure prose. "One can ..., A hundred can ..." and especially the passage I bolded - I would never have guessed that a bot could do it so well. It can do shifts in mood, too. It pivots from a more academic style of speech in the first paragraph there to a more intense, emotional mood in the second, and I don't think that's an accident. A lot of human authors could learn from this.

But looking more closely:

The deadliest magical creature that exists—is a wizard or a witch who has decided they have nothing left to lose. A muggle with a gun can shoot down a protected wizard like a duck in a fairground. There is no defence. The only security against a wizard or witch who has decided they have nothing left to lose is to never let them lose in the first place."

The second sentence here is out of place. It would fit nicely with the third, but in the context of this paragraph it doesn't belong - we're talking about the dangers of a wizard with nothing left to lose, but here we're digressing about how easily a wizard can be defeated by a non-wizard, which kind of contradicts the message about how dangerous a wizard can be.

Why is it there?

There is a fair bit of online discussion about whether "a muggle with a gun" could beat a wizard. Supposedly based on a remark by Rowling; the original remark may well be apocryphal but that doesn't matter, GPT-3 has been trained on discussions like the following:

https://scifi.stackexchange.com/questions/160098/did-j-k-rowling-say-that-a-gun-would-beat-a-wand

So it knows that "muggle with a gun" is a phrase that sometimes comes up in discussions involving wizards and conflict, and it knows enough of structure to fit that in, in a way that meshes with the next sentence. But it doesn't have enough sense of meaning to see how it contradicts the speaker's message rather than supporting it. A human writer would never do this.

Another excerpt:

“Now,” Professor Quirrell said, “it is time for class to begin. Take out your first-year Defence textbooks and turn to page thirty-two.”

There was a ragged scraping of chairs and pages as most of the class hastily obeyed. A few, Harry noticed, though neither Anthony nor Tracey, remained in their seats, with dazed expressions on their faces.

This makes no sense: if "most obeyed" the instruction to get out their text books, why have only "a few... remained in their seats"?

GPT-3 understands that "most... a few" is a common construction in this kind of writing. It understands that scraping of chairs is associated with seats, and that in this kind of story students are often leaving or remaining in their seats. But it doesn't understand that reading a textbook and leaving one's seat don't fit well together.

Professor Quirrell looked at his watch with a frown. “I had originally planned to spend another three minutes talking about something completely different,” the man said. “But I can see that our precious class time has run out, so you shall instead spend the rest of the class period completing your first essay. Your assignment is as follows. On this piece of parchment,” Professor Quirrell tapped the blank piece of paper in front of him, “I am writing a short paragraph describing one fact about the nature of mountains.” There was the sound of scratching quills as the students began to write. “At the end of the class period, you shall each read one of your fellows’ essays and say out loud why you did not award full marks. You will then write your own essay, explaining in as much detail as possible why you would have given a higher mark. Your essay shall run no less than eight hundred words.”

There was silence in the classroom as the students worked.

“I am looking forward,” said Professor Quirrell, “to grading your essays. Now who has read my description of the mountain? Raise your hand, and be honest—in matters of this sort I do not play silly buggering games of honesty like that of Professor Snape.”

The hands of the Slytherin students went up, those of the Ravenclaws next, and last the Gryffindors; only Harry kept his hand down.

Professor Quirrell surveyed the class with no expression on his face, as Hermione’s hand rose into view.

“I can see no reason at all why you should be assigned any marks, Miss Granger, as your essay is objectively incorrect...

Here, he tells them that the class time has run out - but then tells them to use the time to complete an essay. After telling them that he's looking forward to grading their essays, he tells Hermione that hers is incorrect, even though there's been no time for her to write it yet.

...and so on. Pick just about any GPT-3 excerpt, read closely, and you'll find glitches like this.

In a short excerpt, those are less noticeable than its impressive capabilities with style and structure. But the longer it goes, the more non-sequitur-y it's going to become and the harder that gets to ignore.

I can see it being used as a helper tool to spice up one's prose, and as a generator of ideas. But it'll need a lot of hand-holding and editing along the way, and I don't think that's something that can be fixed merely by throwing more processing power at it because it's not designed in a way that allows it to learn logical consistency.

Now, will they be great literature? At present capabilities, probably not. But are Lit stories or Amazon shorts great literature? I think it would be feasible to train an AI program to turn out competently done erotica stories (other genres as well).

If "competently done" includes "has a coherent plot", this is going to require major advances.

 

[windar]

Sure AI has a way to go. But you've picked a work that has a complexity beyond the typical Lit story and a very small sample size. I honestly feel the program would do considerably better at imitating a basic stroke story of which there are a few hundred thousand data points. And also fewer readers would care about the plot deficiencies if it had the right stroke elements.

But the only way to really know is to do the experiment, which I don't propose to do since a successful result would put me out of a job (well, not really, but a modest source of income and of some enjoyment).

 

[Bramblethorn]

Sure AI has a way to go. But you've picked a work that has a complexity beyond the typical Lit story and a very small sample size. I honestly feel the program would do considerably better at imitating a basic stroke story of which there are a few hundred thousand data points.

Erm, "small sample size"?

There are on the order of a million Harry Potter fanfic stories available on the web. Fanfiction.net alone has over 800,000 of them, Ao3 over 340k (probably some duplication between the two). So very roughly twice the 500k stories available on Literotica. I'm not sure exactly what pages are included in the Common Crawl database that GPT-3 uses as the major part of its training corpus, but I'd expect it includes a fair chunk of that fanfiction.

(It pretty much has to; there's no way it'd be able to write HP pastiche as well as it does if it only had the original books to work from.)

The logical inconsistencies in those excerpts aren't a matter of sample size, and nor is it story complexity - it's managing to contradict itself even within a couple of paragraphs. They come from the fact that "logical consistency" simply isn't a thing that GPT-3 is designed to achieve. This isn't a problem that can be solved by increasing the size of the training corpus.

Not to say it couldn't be used to help an author writing in any genre; [url="https://unokay.com/blog/using-gpt-3-to-write-pornography]this snippet[/url] gives an example of using it in what looks like the lead-in to a porn scenario. But it'd be a very interactive process, with the human partner needing to take care of things the AI can't.

 

[windar]

The logical inconsistencies in those excerpts aren't a matter of sample size, and nor is it story complexity - it's managing to contradict itself even within a couple of paragraphs. They come from the fact that "logical consistency" simply isn't a thing that GPT-3 is designed to achieve. This isn't a problem that can be solved by increasing the size of the training corpus.

I never said that the AI would necessarily produce a story that was logically consistent. I said it could produce a Literotica story...

 

[SimonDoom]

I never said that the AI would necessarily produce a story that was logically consistent. I said it could produce a Literotica story...

To paraphrase Oliver Wendell Holmes, the life of an erotic story is not logic, it's experience.

One can debate in the abstract all one wants. I'd like to see an actual example of a 5,000+ word AI-generated story that would work as a Literotica story. If there is one, I'll believe it. But from what I've seen so far, I think Bramblethorn is right. The story might be grammatically correct, but it likely will not have the continuity and coherence in plot and character to be a decent story.

I'm sure AI will get there eventually. Maybe sooner rather than later. I'm not sure it's there yet.

 

[lovecraft68]

I'll keep my response very simple. Between first cousins its technically incest. But in the realm of the taboo category and genre its as weak as it gets and its not as popular as more direct couplings like siblings, parents/adult son or daughter. I'd rate it below aunt/uncle as well even though its about the same 'distance' wise, there's less taboo it seems.

 

[SimonDoom]

I'll keep my response very simple. Between first cousins its technically incest. But in the realm of the taboo category and genre its as weak as it gets and its not as popular as more direct couplings like siblings, parents/adult son or daughter. I'd rate it below aunt/uncle as well even though its about the same 'distance' wise, there's less taboo it seems.

Agreed.

Sometimes, on the subject of story categorization, people get too worked up about the "logic" of the categorization, or about how the categorization would play out in the real world. None of that is relevant. All that matters is the degree to which the categorization assists readers in finding the stories they want.

Cousin sex technically is incest, but the taboo element is weak, so it lacks the sizzle of more intimate incest. That's why I'll never write a cousin story. From my point of view, there's just no erotic interest.

 

[lovecraft68]

Agreed.

Sometimes, on the subject of story categorization, people get too worked up about the "logic" of the categorization, or about how the categorization would play out in the real world. None of that is relevant. All that matters is the degree to which the categorization assists readers in finding the stories they want.

Cousin sex technically is incest, but the taboo element is weak, so it lacks the sizzle of more intimate incest. That's why I'll never write a cousin story. From my point of view, there's just no erotic interest.

I think the reason aunt/uncle plays out better even though its 'extended incest' is there is also the age dynamic, an aunt is a milf, the uncle an older man so the combination of age gap and family gives it a punch cousins don't have.