bot writing stories III

[astuffedshirt_perv]

Rather than resurrect the old thread, here we are 7 months later with a newer iteration of writing bots.

The previous thread discussed GPT-2 and its release. I have gotten GPT-2 to run on my computer and have generated clear, grammatically correct prose (in all three releases, though OpenAI did not release the full Mega model). The problem with using it to write a story is that it is like a writing Pachinko machine, and it just wanders away. The prose can be good, but at only max ~800 words an iteration it will end up God knows where.

The idea of GPT-2 and others (like Grover) is that it 'predicts' the next word, so I got to wondering if it could predict the previous word. And if so, could you give it a first sentence and a last sentence and have it generate grammatically correct words between the two. And now AI21 had put HAIM on the web, which does exactly that.
Link to HAIM. I put in the bolded lines, got back the rest (settings: custom, medium).

Erdor drew his blade and braced himself against the mighty Galforang.
Erdor's blade cut through the galforang, killing it instantly if it had been a human. Erdor's blades cut through the body of the Galforang in a similar manner, leaving the dead creature's flesh hanging out and its bones crushed beyond recognition. The dead Galforang's flesh and bones spilled out into the desert, leaving only the bloodied remains of the galforang's body and its head. The galforang's flesh was still warm enough to touch, but it was a very cold and very dark place.
Erdor stepped back from the dead Galforang, its blue blood seeping into the ground beneath it.

AI21 has not released the code, it is just a webpage right now, but surely this kind of AI writing is coming fast.

 

[BelleCanzuto]

I'm not worried about robot overlords taking this over anytime soon.
I tried it, using the first and penultimate lines of a story i just submitted.
And I got this:

Symon woke himself up laughing. He was still smiling, and now he knew that he was also alive. Symon woke again, and stood in the hall, looking out the window through the open windows. A woman had come into the room, sat on the floor, got in bed with the other woman. She was a middle-aged woman of about thirty-nine and slender. She was wearing a blue dress that was slightly too short for her thin frame, though she could make out everything and did. "I'm Anna's sister," she said. Symon jumped up and walked home, but then his grandmother made him do it again.

She wrote, "start here, you missed a spot."

Suffice to say my story is a little different.
Honestly, this gibberish just makes me wonder what universe of samples the algorithm is drawing from.

 

[BelleCanzuto]

Some classics, by HAIM

Moby Dick -

Call me Ishmael. Now it is time I tell thee my name. I am Ishmael. I am on this earth in God's hands. I am only escaped to tell thee the reason why I am on this earth. I am Ishmael. In God's hands I am. And when thou dost speak to me, not only will I return that voice of mine in his ear, but all men as well hear it as if in my spirit, who have not heard thy voice in man's. My name is Ishmael. Call me Ishmael. Now it is time I tell thee my name.
And I only am escaped alone to tell thee.

A Tale of Two Cities -

It was the best of times, it was the worst of times. It was just after World War I when I was still a student in my native New Jersey. My first year of life as a soldier in the army had been the very best of times, the best of times I have ever known. The country and me was beautiful and it was so easy to go to the places that we went to, the places that the American people had visited for generations. I was with my best friend in one beautiful place we went to so many times, and it was wonderful and we took in many wonderful people. It was my most wonderful year of life. Never has the world been any better.
It is a far, far better thing that I do, than I have ever done; it is a far, far better rest I go to than I have ever known.

{{I could do this all day}}

Fahrenheit 451 -

It was a pleasure to burn. "I'm sorry," said Vah'az. "But I need to speak to you." There was no sound. The silence was silent as night. The stars danced slowly, their light dancing between the stars they had seen, illuminating the night to them. "The Night of Dusk," said Vah'az. "I hope you'll allow me to tell your story." "I'm glad it is this way," said the boy. "No worries," said Vah'az. "This will lead to me being taken away and taken to the City." The boy nodded.
"When we reach the city."

{{ok, I'll stop. also, the last lines of these novels are as generated by a search engine. I didn't personally double check. So if they're wrong, (or if I misremembered the opening lines) don't blame HAIM. Really, don't blame HAIM for me being bitchy about what are probably legitimate improvements in automated writing. It's just. Well, these samples are absurd. And sometimes I'm a bitch.}}

 

[astuffedshirt_perv]

Moby Dick -

Call me Ishmael. Now it is time I tell thee my name. I am Ishmael. I am on this earth in God's hands. I am only escaped to tell thee the reason why I am on this earth. I am Ishmael. In God's hands I am. And when thou dost speak to me, not only will I return that voice of mine in his ear, but all men as well hear it as if in my spirit, who have not heard thy voice in man's. My name is Ishmael. Call me Ishmael. Now it is time I tell thee my name.
And I only am escaped alone to tell thee.

Yeah, putting in the first and last lines of a novel would be pretty pointless. HAIM is generating a paragraph. Put in the first and last lines of a paragraph.

 

[BelleCanzuto]

Classics by HAIM, take two

Moby Dick -

Call me Ishmael. I am a sailor in the Royal Navy. I like the ocean far, far away. I love the sound of the seas, and the beautiful water, which rises and falls with the sea. I like the idea of the sea being a mother to my children and me. I can hear the ocean at sea when it's full in spring, and when it's full in the autumn. The ocean's a father and a mother when I have a child. Now, I am a sailor at home in a ship, but I feel very much the affection of men of sea. If they but knew it, almost all men in their degree, some time or other, cherish very nearly the same feelings towards the ocean with me.


Fahrenheti 451 -

It was a pleasure to burn. But if we had had the day to set it free, it would have fallen again. I'd have had a good morning in." It was one of those rare, out-of-character fires, but they all worked. He set the books down and the fireplace turned on on dry white fire. The books fell out. The fireplace was filled with black ash. Bolero opened his eyes and breathed. He sat and opened his mouth and opened his nose. It was beautiful. "Do you understand me?" He couldn't say what he heard the fire say. While the books went up in sparkling whirls and the wind turned dark with burning.

Ok. I'll give the bot the Moby Dick paragraph. That one at least sort of makes sense and keeps the idea of an emotional relationship with the ocean. But the Fahrenheit 451 paragraph is clearly random sentences with the barest relationship to two of the words in the given sentences.

I realize that software like this is already being used for some commercial writing. I've seen stories about bots writing sports stories, and other short items when the purpose is to convey easily digested, factual information. But I have an aversion to assertions that such obviously infant technology is going to soon be taking over from humans.

 

[whowrotethisshit]

I mean, it's certainly getting better and sure people would read it but, to me, it aint got that zip. It has never fell in weird love and had weird two a.m cigarettes and coffee because of it, it doesn't have a great-aunt who stole collection plate money, it's never been kicked in the teeth or spit blood onto its new shoes. Warm to the touch or not. It ain't got that zip.

I'm broke before robots, I'ma be broke after robots. To quote the CBGB movie,
"Seven dollars? How am I supposed to live on this?"
"Hey, man, art sucks."

Moby Dick -

Call me Ishmael. Now it is time I tell thee my name. I am Ishmael. I am on this earth in God's hands. I am only escaped to tell thee the reason why I am on this earth. I am Ishmael. In God's hands I am. And when thou dost speak to me, not only will I return that voice of mine in his ear, but all men as well hear it as if in my spirit, who have not heard thy voice in man's. My name is Ishmael. Call me Ishmael. Now it is time I tell thee my name.

Honestly, this is me when i get pulled over and I remember I forgot to throw away all my expired insurance print-offs and can't find the newest one.

 

[Bramblethorn]

These AIs are certainly getting better at mimicking sentence structure and grammar, even some larger-level structure. In the first example he hits it, it's dead, it's "still warm", its blood is seeping into the ground.

But by my understanding, it doesn't have any internal representation of the facts it's describing. It's basically a very sophisticated version of the Markov-chain game where you start writing a sentence on your phone and then let predictive text complete it.

In the first example, for instance, the AI has been trained on a corpus where phrases like "he drew his sword" are often followed by concepts like "killing" and "killing" is followed by mentioning of bodies being "still warm" and blood "seeping into the ground".

If you search on "cut through the" and "killing it instantly" you'll find a bunch of examples where those phrases show up in close proximity - some discussing animal slaughter, others swords-and-sorcery etc. So HAIM has figured out that it can put those together, and in that first bot-generated sentence it works.

Searching on "instantly if it had been a" gets several hits, and likewise "if it had been a human". So the bot can put those together to get "killing it instantly if it had been a human".

(Note that this doesn't actually make grammatical sense, in context - it would need to be something like "deep enough to have killed it instantly if it had been a human".)

The second bot sentence seems to partly repeat and partly contradict the first one. The first one told us that he had "a blade", the second that he has "blades". The first said that he didn't instantly kill it, the second seems to imply that it's dead when he hits it (since it's now "the body of the Galforang"). What's going on here?

Well, there are several texts where "cut through the" and "in a similar manner" appear in close proximity. They're generally not describing fantasy combat; they're things like dissection or construction instructions. You cut through something, it tells you how, and then you cut through another thing in a specific manner. But the "cut through" bit overlaps with the language you'd find in a swords-and-sorcery story, so HAIM has tried to fit that structure in here, where it doesn't really fit.

And it doesn't know that the Galforang isn't dead. As far as it's concerned, it just wrote "Erdor's blade cut through the galforang, killing it instantly..." and that's the kind of sentence that can reasonably be followed by stuff about "the body of the Galforang" and blood spilling out, yada yada. It doesn't understand that "if it had been a human" changed the meaning of that text.

Similarly: "its bones crushed beyond recognition"? "Bones crushed" is fine. "Crushed beyond recognition" is a common phrase. So it's glued them together to get something that doesn't make much sense here.

In a few paragraphs, something like HAIM can get away with not understanding what it's actually talking about. But the longer it goes on, the more visible those limitations become. Outside of things with very standardised forms, writing a coherent story that passes the Turing test is going to require something very different to this, probably something which actually has an internal representation of the events it's describing and not just "these kind of words often go together". (In machine learning terms, maybe looking at something like latent feature methods?)

Good enough for political speechwriting, though.

 

[jehoram]

Good enough for political speechwriting, though.

Particularly for some politicians I could think of. I don't want to drag this discussion into something that would fit better on the General Board or Politics Board, but when I hear Trump talking about something that he doesn't really know anything about, I remember my school days, when I would hear a kid bullshitting his way through a book report that he hasn't read: "The book is Treasure Island. It is a book about treasure, which is something that is good and a lot of people want. And it has islands in it, which are important if you're sailing around and you don't want to sail anymore. The treasure is on an island. That's what the book is about. It's a very good book. A classic, some would say. And, as I said, because this treasure is on this particular island, people have to go there if they want to get the treasure. That's what the book is about."

In fact GPT-2 would probably do a better job.

 

[BelleCanzuto]

Particularly for some politicians I could think of. I don't want to drag this discussion into something that would fit better on the General Board or Politics Board, but when I hear Trump talking about something..."

~snip~

In fact GPT-2 would probably do a better job.

I had the same thoughts. But yeah, politics...

 

[astuffedshirt_perv]

These AIs are certainly getting better at mimicking sentence structure and grammar, even some larger-level structure. <snip>
In a few paragraphs, something like HAIM can get away with not understanding what it's actually talking about. But the longer it goes on, the more visible those limitations become.

Yes. Assuming this code or similar code is released on, say, Github (seems likely), you could take a text heavy story you like (won't really work for dialog heavy stories), strip out the first and last sentence of every paragraph (trivial with MS Word/Excel) and write a simple script to feed those lines into the code one after the other. Voila, you have a new(ish) story.

Or write a detailed outline for your new story, comprising the the first and last sentence of paragraphs. Feed that into the generator.

Or just write the first and last sentence of your story, generate, write a sentence a few more alone, generate.

Say you can compose 1000 words in a sitting. If those words are then expanded by generating an additional 2000 words, your writing is getting turbocharged. Don't like the generated words? Edit it or run the code again. Plus all the new words will be spelled correctly and grammatically correct.

 

[SimonDoom]

I find this issue very interesting, but I don't have a good understanding of how this software works. Is there a good article that summarizes it? One that is not too technically challenging? I have only the most basic exposure to and knowledge of coding, so I can get confused by the terminology.

Is the writing algorithm designed only to figure out what the next best word is, or does it "look ahead" in a meaningful sense?

I have some ideas for some erotic stories incorporating AI concepts, but without a better understanding of those concepts, writing the stories is a challenge.

 

[Bramblethorn]

I find this issue very interesting, but I don't have a good understanding of how this software works. Is there a good article that summarizes it? One that is not too technically challenging? I have only the most basic exposure to and knowledge of coding, so I can get confused by the terminology.

Is the writing algorithm designed only to figure out what the next best word is, or does it "look ahead" in a meaningful sense?

I have some ideas for some erotic stories incorporating AI concepts, but without a better understanding of those concepts, writing the stories is a challenge.

This is my rough still-sleepy interpretation of stuff outside my area of expertise (I have some machine learning knowledge, not in this area) so I'm doing my best but I don't promise to be 100% accurate.

I'm drawing on these posts:

https://www.ai21.com/haim-post
http://jalammar.github.io/illustrated-transformer/
http://jalammar.github.io/illustrated-gpt2/

First important concept is neural networks, which are a big part of modern AI.

Let's suppose you want to train a computer to recognise images of elephants. Recognising an elephant all in one go is a tricky task, so let's break it into stages.

You might start by training a function (I'm going to call it a "layer" for reasons that will soon be apparent) that recognises specific shapes over a small area - line, curve, dot, etc.

Then you might build a second layer that receives the input of that first layer and outputs higher-level features: smooth curve, wrinkled vertical-ish line, etc. etc.

Then another layer to translate them into higher-level still: two smooth curves almost parallel, meeting at a point, is a tusk. Two curves with a dot in the middle is an eye. A bendy wrinkly thing is a trunk, two vertical wrinkly lines plus some pobbly bits at the bottom is a leg, etc. etc.

...and so on until you get up to a layer that says "this looks like an elephant" (or not).

One way you could do this is by building each of those layers individually and optimising it individually. You take a bunch of training images classified as "line", "curve", "dot", etc. and you optimise your first-layer function for classifying those. Then you take a bunch of "line, curve, dot" data and optimise the second-layer function to classify that into "smooth curve, wrinkled vertical" etc. And so on.

Problem is, (a) this gets immensely time-consuming really fast, and (b) it's hard to know exactly what kind of features might be important. So instead of micro-managing what the individual layers do, you might just do it like this:

- input the raw pixel data (say, a million pixels)
- have a million-odd functions that each input raw pixel data and output a number
- have a second layer of functions that input numbers from the first layer and output their own numbers
- stack these up in as many layers as you want
- final layer outputs just one number between 0 and 1, its prediction for "is this an elephant?"

And instead of explicitly defining what each of those functions should be, you let the computer figure it out. You define a general form for each function (logistic functions are commonly used) and then, within the framework imposed by that general form, you optimise to find functions that, when put together, do a good job at predicting "elephant".

By building those up in layers, you give this network the capability to build simpler entities into more complex entities. It almost certainly won't do it the same way a human does, and if you're not very careful you can end up training it on the wrong cues (there's a famous example of an AI that was supposed to learn the difference between cancerous and noncancerous moles, but the training photos included rulers next to all the cancerous moles, so it just learned to detect rulers...)

But with careful training, and with enough computational power, neural nets can be quite effective at certain tasks. "Deep neural nets", which are big in machine learning these days, may have hundreds of layers.

Okay, so that's neural nets in general. Language in particular.

You can break speech down into a bunch of "tokens" - individual words, punctuation, etc. etc.

You can then train up a neural net on a large corpus of text (think "download all the writing you can easily get in electronic form") and teach it something like "given the previous ten tokens, predict the next one". (Rather than just giving you one possible answer, it'll give you a bunch of possibilities and their estimated probabilities, so you can randomly select from plausible options instead of taking the most likely option every time.)

For instance, if the last token was the word "I", the next token is very unlikely to be a full stop, but things like "am", "will", "was" are much more plausible. Earlier tokens will help identify the tense, etc. etc.

By my understanding - and this is where I'm less sure of myself - the Transformer approach that HAIM and GPT-2 is based on a layered approach to prediction. At the bottom level are the input tokens e.g. "previous 10 words", then it builds those up into higher-level entities before coming back down to predict the next word. I think this where it gets the capability to recognise things like the structure of a persuasive essay and the overall style of a passage - it's not just "what type of word follows these words?" but also "what type of sentence follows these sentences?" and even "what kind of paragraph follows these?"

The GPT-2 approach took a user-supplied beginning and continued from there, which often led to it wandering a long way off topic. HAIM has been tweaked so it looks at both the beginning and the end, and writes something to fit in between those two.

 

[Bramblethorn]

Out of curiosity, I tried it on a few of my own stories.

By the time I found the place I was pretty sure going to the work Christmas party had been a bad idea. The store was quite large and had a lot more merchandise, which probably didn't make up for the extra storage space or even the fact that I couldn't fit a single pair of pants in the place. The second-floor walk-up upstairs had a great view above the street below and there were a lot of places to eat in the parking lot. I guess I'd like to think I was in a shopping district with lots of coffee shops like these: The best part was that my kids got to experience the magic of downtown Brooklyn all in one place and I thought I'd give it a shot myself. I was wrong.

And after I'd outed myself a couple of weeks ago...

Seems like it's riffing mostly off "found the place" there, ignoring the Christmas party bit and the final sentence.

The "I thought... I was wrong" seems like a call-back to the "bad idea" - like the GPK-2 examples we discussed a while back, it clearly does have some ability to learn structures.

The "...had a lot more merchandise, which probably didn't make up for the extra storage space" is a weakness that I've noticed in a few samples from these generators. Grammatically speaking, this structure is fine, but in normal English we'd only use it when the first item is positive (check) and the second is negative (not-check).

No doubt he'd saved money by holding the party at his place instead of booking a restaurant (there was probably a tax deduction involved somewhere in that) but when I walked through the door I could see he hadn't stinted on the catering. They weren't the cheapest meals, but they'd been well received: There were plenty of appetizers and nasi lemak, a vegetarian version, and some fried egg on the side. A generous selection of cheeses were the order of choice, especially with prices falling this season. The main attraction of the meal came from the bar, where I encountered a cast of characters I had never heard of. In the middle was Peter Quill (Chris Pratt); in the end, there was a group of teenagers (who were all teenagers in an ironic way) who were in a bar with me.

Janelle — Peter and RJ's executive assistant — lifted a glass in my general direction and offered a slightly sozzled "Happy Christmas!"

"They weren't the cheapest meals, but they'd been well received" - again, this is a pretty common structure but it doesn't work with those particular statements.

(also, thank you HAIM for that rather unexpected Guardians of the Galaxy crossover)

I nodded. "You're not the type for real estate, anyway."

"Hey! My roomie just went with me." Anna giggled. "He's never made out with me before, is he? You and Sven had a very good time." Elsa smirked. "Yep, he and Sven just have a really friendly bond." Anna chuckled. "And so the date!" Elsa said, smiling. "Oh, my, my!" Anna squealed. And, after giving me an exasperated look, she added, "Come on, girls, why don't you come with us?"

At that point one of the caterers started ringing a bell. "Ladies and gentlemen! If you'd like to line up, dinner is ready!"

"He's never made out with me before, is he?" - suspect it got confused here with "he is" vs "he has".

I headed along a corridor, expecting to find a bathroom where I could lock myself in, but I'd taken a wrong turn. Suddenly, and I can't explain it because I'd already experienced the most terrifying thing since the beginning of this piece, my room had been bathed in flames. My bed looked like it had been burned to the rafters but I felt very safe and could feel nothing but warmth pouring from my body. I don't remember why, but I did feel a great burning sensation over my body. I could hardly move, though, and I felt a heavy burning as though I'd stepped on something. I tried to get away from the flame, but I couldn't go back inside.

I leant against the wall, head on my forearm and cried for a bit.

...well, that was a bit melodramatic.

I touched her neck with my lips, breathing in the scent of her hair, and she sighed very gently. "I'm glad you've found someone this warm." "You mean it's a pleasure?" I asked, flicking my lip and looking at her. It looked to be a pretty good approximation of what I thought we might have here, but it was very warm. "You're nice. I guess it's nice to have somebody who's warm to you." "I hope you find someone who's kind," she said with a light smile, her words almost perfect and perfectly in tune in their phrasing. "Yeah... no way." The evening had been incredibly romantic for a while, but the evening had changed everything.

My nails scraped over the fabric, sending sensations through to the warm body beneath, and I leaned forward to kiss her just below the ear.

I am officially No Longer The Worst at sexy dialogue.

Hello everybody. I'm Ashley, your host and facilitator today. You've met my assistant Cherie, and our driver is Derek. In a moment I'll tell you some more about what we have planned, but first I want to set a few ground rules. Number one, I'll be honest with you: I don't want to have anything to say about what you want or what you need. You are a free person and it's not your job to define what you want or what you need. You're allowed to do what you want. You've probably been doing this for a while and it's the best thing that ever happened. You love being healthy. You love meeting new people. You love getting together to drink some quality alcohol. You love meeting new people. You love your family. That's what we're here for because you need to grow.

Rule number two, you can say no to any activity. Today is about pushing your boundaries and discovering what you're capable of, and what makes it meaningful is that you choose to do it. We'd like you to stay with us for the whole day, but we're not going to push anybody further than they're willing to go.

I was curious about whether it would deduce from "Rule number two" that there ought to be a rule number one somewhere in there, and yes, it did.

Kelly had her hand between Sigrid's shoulders, so I moved mine down, stroking her hips. She seemed to like that; I could feel that she was concentrating on Kelly just then, but she squeezed my arm as if to say: more please. "Just a sec, little girl," she said softly, then turned away and turned back toward the desk. I could see the corners of my mouth and nose turning a little purple. The taste of her fingers ran along my tongue. "You're right, your sister's hot. I want you to love her that way. To make it right. I don't know if we should be a couple. I don't know if we're ready yet." I leaned forward, and her neck began moving a little further. She grinned, and I noticed she was starting to get big.

As it was, it took much of my willpower to keep my erection in check.

Her fingers stroked my belly, pushed down just under my belt, and I flinched as they found the head of my cock. She had her tongue down my throat, and she was sucking down my pussy like a child. It was a little uncomfortable, but I was undeterred. After a while, my pulse started to slow and I felt a little nauseous, but I couldn't leave. Her legs were still on my ass, she was so tight that she couldn't pull away, and I began to hear her orgasm in my head, and it was so amazing. She had gone so far as to hold my cock in her mouth for a long time. It felt like she was holding a giant snake in her mouth, twisting.

Then she encircled me, squeezed me, slowly pulled down, and up, and down. It felt fantastic.

...okay, who's been training HAIM on futa stories?

 

[ElectricBlue]

She had her tongue down my throat, and she was sucking down my pussy like a child. It was a little uncomfortable, but I was undeterred. After a while, my pulse started to slow and I felt a little nauseous, but I couldn't leave. Her legs were still on my ass, she was so tight that she couldn't pull away, and I began to hear her orgasm in my head, and it was so amazing. She had gone so far as to hold my cock in her mouth for a long time. It felt like she was holding a giant snake in her mouth, twisting.

What the considerable fuck? Your AI has conjured up a creature with both a pussy and a cock (a giant twisting snake, what's more, lol) and look at the repetition! Needs a good editor and Anatomy 101.

And a lot more work. I am not threatened by this .

 

[Hypoxia]

@Bramblethorn: When will you submit that?

 

[Bramblethorn]

What the considerable fuck? Your AI has conjured up a creature with both a pussy and a cock (a giant twisting snake, what's more, lol) and look at the repetition! Needs a good editor and Anatomy 101.

And a lot more work. I am not threatened by this .

I mean, "woman with pussy and cock" is a moderately common fetish. Pretty sure AwkwardMD has at least one in that vein. But "tongue down my throat" and "sucking down my pussy" plus "legs on my ass" all at the same time takes some doing.

(I did write one story where the femme fatale had her lips on the guy's dick and her tongue in his ear simultaneously, but that was his clue that something had gone very very badly wrong.)

@Bramblethorn: When will you submit that?

...you know, I feel like this could be the basis of a fun Author Challenge contest. Generate a few paragraphs of HAIM-porn, and then challenge authors to write a story around that. Bonus points if it fits seamlessly into the finished work.

Experimenting a bit more:

"Be gentle, it's my first time," she whispered. With that, the other two were dragged off into a large room, with several walls of a green and gold lattice. The one woman in the room, the one who started to cry, took a step towards Teferi and kissed her cheek. Teferi had tears pouring down her face as they parted, and she turned her eyes back to the other two. "What did you say?" she asked them, her voice a whisper against the darkness. The one woman's face was pale, scarred, and pale red. When the woman looked at her, she knew it was the words of an old, fallen warrior. He had lost a hand in war. He was bound and bound, and they would not return to him until their fate was complete. They were not. The woman bowed her head down, and closed her eyes. The other two watched as the others watched around her. Her expression, as bright as the moonlight, seemed almost too radiant to be real to Teferi. Her eyes were closed completely, and her mouth was still covered by hair. She had not said a word.

One does not simply walk into Mordor.

Ra ra Rasputin, lover of the Russian Queen. I love that it's an image of a man and a woman playing cards. I love the silhouette of someone looking directly into the camera. What a great image to have to show our readers! The image is not one of love, but rather lust ... In many ways this is a true portrait of an attractive young Russian female. We see clearly that she is young, but also that she is beautiful in a way that a young man cannot be. In a way the woman's face is as if she is one of a child, as if there were a doll-like figure in front of her that's holding her hand. The image doesn't only represent love on the one hand, but also a desire to be liked. It may not be love, but it is love. The girl is clearly young and attractive, like in a fairy tale, though the portrait doesn't show it. This is in fact where some people think we would have seen a young girl, with hair long, but she is not quite mature, or tall, unlike the man on the opposite right.

You spin me right round, baby, right round.

Dear Penthouse, I never thought this would happen to me, but last summer you sent me to my third birthday party at the Playboy Mansion. One of the party guests was my sister, Jody, who lives on Maui. And Jody has had a long and happy life; she has given me a steady job and two jobs in the adult industry. Jody, it should go without saying, is a nice person and a devoted wife. I never thought that at this time in my life, after having gone through so much turmoil and struggle, my children would be so happy and secure. For now, I am so grateful to them for their time with me. But I will never, ever have a second chance at that moment again. Not for myself, anyway. I will be working full-time during the summer of 2008 and I will be trying hard to manage my busy household life and stay in shape. But in reality, I have a lot more to give. I will be at my third birthday party this coming summer. You have been kind enough to send me. Your birthday party is an amazing night of happiness and fun. I wish to share my story with you.

"It is a far, far better thing that I do, than I have ever done; it is a far, far better rest that I go to than I have ever known."

well THAT one's not getting past Laurel's filters.

"Say, is that a tentacle?" Anna asked and her eyes widened, thinking, feeling what she thought she saw as an actual tentacle. "It's my favorite," Elsa admitted simply. The woman gave her a curt nod and Elsa nodded her approval. "What about?" she asked with a grin. "A tentacle doesn't get in the way of my view." Elsa smiled at her, as Anna took back her hand and made it move. The woman gave her a nervous look before getting up and backing out. She began to move slowly towards her, watching with her other hand. Anna was a good distance away, and she took her time to react as well. Elsa began to wonder what Anna's intentions were. What could she possibly do before she found Elsa's tentacle and started to move it? Did she have any idea of what her girlfriend wanted? Could she even give this away? Well this would be a huge surprise. As Elsa looked at Anna, she looked over her shoulder, wondering if any of what she was witnessing was real. The woman reached out her hand and Elsa let her take it off before she could object. The tentacle hit her face first and fell onto the grass.

She shuddered in orgasm and collapsed, completely spent.

Would read.

Without breaking eye contact, she slowly stripped off her clothes one piece at a time. A little while later she grabbed her friend's hands and said, "I'm gonna go in and have a talk with my mom." And the girls walked out. Sasha was the first who came back. After some time she brought her friend's hand and said, "So what was the plan? I think she knew where I was going." I sat in a back room of a hotel, under a huge banner. At first it didn't make any sense, but after a while I got a handle on how this could have happened. I looked out the window and saw that this other town was surrounded by two large cities. And I saw what looked like an army of some sort, surrounded by military equipment and large tanks. I thought I might have escaped with Sasha, but just then two officers escorted me away. They were the first to leave and they were all dressed in white and had white clothes. Once they stepped outside a few police officers followed them, which is normal. And the military equipment and tanks and everything just came out of nowhere. The officer who escorted my friend to the airport told me to get my suitcase because he had some questions. The officer asked me a few questions.

It was the most fun I've ever had at a PTA meeting.

...okay, I'll stop now.

I don't know that I'd rely on it to write the whole story, but this looks like a promising source of bizarre prompts.

 

[astuffedshirt_perv]

I find this issue very interesting, but I don't have a good understanding of how this software works. Is there a good article that summarizes it?

I think the original GPT-2 blog post is pretty good. And it is only from February of this year! Also relevant to note that they did not release the full version, though others have done the same thing and have released a full version.

"Our model, called GPT-2 (a successor to GPT), was trained simply to predict the next word in 40GB of Internet text."

Imagine you can retain every article you have ever read in your head in perfect order. Now imagine you come across this sentence fragment:

Alice was surprised to open the door and see

What is the next word? Based on all the articles you have read, you can make guesses. Maybe it is a "a" or "the" or "Dad". Probably not "briquet". Then repeat.

Note that the "40GB of Internet text" is not heavy in erotic stories, if at all, but the model can be trained. You could download all of asstr (or literotica for that matter) and 'train' the model to guess the next word from an erotic database.

 

[jehoram]

I find this issue very interesting, but I don't have a good understanding of how this software works. Is there a good article that summarizes it?

I learned about GPT-2 from this article in the New Yorker:

https://www.newyorker.com/magazine/2019/10/14/can-a-machine-learn-to-write-for-the-new-yorker

 

[KeithD]

This is completely out of my concept of what a writer is and does.

 

[SimonDoom]

This is completely out of my concept of what a writer is and does.

But that's what makes this whole idea so interesting. Can a machine, doing something completely different from what you do, nevertheless fool people into thinking that it IS doing what you do? It's an unsettling idea.

 

[KeithD]

But that's what makes this whole idea so interesting. Can a machine, doing something completely different from what you do, nevertheless fool people into thinking that it IS doing what you do? It's an unsettling idea.

Not really. Not to me. If people want to read BOT-generated stories, let them (same as if they want to read E.L. James, let them). There are plenty of readers to go around. Just keep my own access to readers open and I'm happy to let them chose what floats their boat. That doesn't mean that anyone who uses extensive external help in writing is a writer. Because they are only a writer to the proportion of their own effort and thought, character and plot construction, and delivery ability has been brought to bear.

 

[BelleCanzuto]

But that's what makes this whole idea so interesting. Can a machine, doing something completely different from what you do, nevertheless fool people into thinking that it IS doing what you do? It's an unsettling idea.

I think it's fair to say, based on the numerous examples in this thread: not currently.

Personally, I doubt it'll be anytime in the next 10 years.

 

[Hypoxia]

This could turn NaNoMo into NaBotNoWk, a novel per week, like Simenon but with time left over for employment and self-abuse. Use LIT archives as the AI's training base. Fun.

 

[CJoshua]

It does sound weird, when you try to read the bot-generated stories on FiftyShadesGenerator (I have just tried and I didn't like it, though it was amusing). The computer-generated stories will always lack sensuality and temperament.



Main Literotica Forums >> Authors' Hangout >> bot writing stories >> Next topic >> Books on how to write and question about submissions with essaytyper

 

[CJoshua]

An article for a quick read > Reid McCarter: Here's a language bot just doing its best to write filthy erotica for humans