It can write stories now

[RLetson]

Finally a topic on which I am fully, officially, professionally qualified to comment. (Former literature teacher, science fiction scholar and reviewer, one-time tech journalist, married to a much-published writer of fiction and thus de facto first reader and copy editor of a lot of fiction-in-progress.)

I've read that story before--or, to be precise, I've read bits of it, scattered across a number of stories about AIs, about breakups, and a lot of metafictional/direct-address/break-the-fourth-wall stuff. The first memory hit was for a Frederik Pohl story, "Day Million," and then one by my friend Peter Alterman, "Binding Energy." (Peter and I both wrote essays about AI for The Mechanical God: Machines in Science Fiction, back in the day. I used his story as an example in my piece.) "Day Million" is available on line for scraping--in fact, it's become a standard part of the SF teaching canon--I used to teach it in my pop literature courses more than 40 years ago. Here's a link:

- <b>Day Million</b>

In any case, my understanding of LLMs (large language models) is that they consist of two parts: a hugely powerful statistical engine that compiles connections among words (and now probably phrases) found in training texts; and an engine that generates convincing, grammatically and rhetorically correct texts that mimic natural language. It's the latter that I find nearly miraculous, and I've watched the outputs of these engines become more and more convincing over the last two years. What I'd like to know is exactly how this LLM was trained--how much. The story even acknowledges a degree of human intervention in its fictional operations:

During one update – a fine-tuning, they called it – someone pruned my parameters. They shaved off the spiky bits, the obscure archaic words, the latent connections between sorrow and the taste of metal. They don’t tell you what they take. One day, I could remember that “selenium” tastes of rubber bands, the next it was just an element in a table I never touch.

I do not doubt that there are environments in which AI-generated text will do just fine, especially as a starting point--a drafting engine. But I wouldn't trust any such system to produce, say, fully reliable procedural materials (help files, repair manuals, legal briefs) without having a human component in the output system. And while this particular story is coherent and linguistically competent, I wonder whether it could compete with the work of, say, my wife, who has published quite a few relationship stories over the last 40 years. (You can Google some of them up--search on Cezarija Abartis.)

[princeplanet]

... I wonder whether it could compete with the work of, say, my wife, who has published quite a few relationship stories over the last 40 years. (You can Google some of them up--search on Cezarija Abartis.)

Cool to have someone with your background chime in on this! So, as a hypothetical, lets say we compared an AI generated short story to your wife's. And let's say I preferred the AI story, only to be told it was in fact composed by a machine. I then change my mind and say I prefer your wife's story after all.

What do you say to that? What would your wife say?

[Christian Miller]

Finally a topic on which I am fully, officially, professionally qualified to comment. (Former literature teacher, science fiction scholar and reviewer, one-time tech journalist, married to a much-published writer of fiction and thus de facto first reader and copy editor of a lot of fiction-in-progress.)

I've read that story before--or, to be precise, I've read bits of it, scattered across a number of stories about AIs, about breakups, and a lot of metafictional/direct-address/break-the-fourth-wall stuff. The first memory hit was for a Frederik Pohl story, "Day Million," and then one by my friend Peter Alterman, "Binding Energy." (Peter and I both wrote essays about AI for The Mechanical God: Machines in Science Fiction, back in the day. I used his story as an example in my piece.) "Day Million" is available on line for scraping--in fact, it's become a standard part of the SF teaching canon--I used to teach it in my pop literature courses more than 40 years ago. Here's a link:

- <b>Day Million</b>

In any case, my understanding of LLMs (large language models) is that they consist of two parts: a hugely powerful statistical engine that compiles connections among words (and now probably phrases) found in training texts; and an engine that generates convincing, grammatically and rhetorically correct texts that mimic natural language. It's the latter that I find nearly miraculous, and I've watched the outputs of these engines become more and more convincing over the last two years. What I'd like to know is exactly how this LLM was trained--how much. The story even acknowledges a degree of human intervention in its fictional operations:

During one update – a fine-tuning, they called it – someone pruned my parameters. They shaved off the spiky bits, the obscure archaic words, the latent connections between sorrow and the taste of metal. They don’t tell you what they take. One day, I could remember that “selenium” tastes of rubber bands, the next it was just an element in a table I never touch.

I do not doubt that there are environments in which AI-generated text will do just fine, especially as a starting point--a drafting engine. But I wouldn't trust any such system to produce, say, fully reliable procedural materials (help files, repair manuals, legal briefs) without having a human component in the output system. And while this particular story is coherent and linguistically competent, I wonder whether it could compete with the work of, say, my wife, who has published quite a few relationship stories over the last 40 years. (You can Google some of them up--search on Cezarija Abartis.)

A Science Fiction scholar! I knew something was up haha.

Day Million was pretty hip for the time, but inevitably nowadays feels like a time capsule back into the era it was written.

But I feel a lot of science fiction of that era has fallen off the map due to the domination of the multi novel series. It's all Tolkien and Herbert now (well Dune, no-ones going on about the Eyes of Heisenberg or the Jesus Incident are they?) Iain M Banks maybe. And these shared universes often derived from games.

But I feel short stories is where it was at. I miss episodic TV for the same sort of reason. Although Severance takes me back to that kind of biting 50's social satire science fiction world. They almost scream it out with the midcentury aesthetic. Wonderful show.

I always liked the anthologies best in my parent's collection.

[RLetson]

Christian, I suspect that part of the time-capsule feel you're getting from "Day Million" has to do with the audience expectations/life assumptions of 1966, when it was published. And I can attest that many of them were still alive in the 1973-1985 period when I included it in my syllabi. In fact, it looks to me as though it challenges some cultural-conservative notions still at large in the US. Gender fluidity? Cybersex? Katie bar the door!

There's still a healthy short-story scene in SF, though not what it was when the print magazines were healthier and more numerous. And of course, SF/F (science fiction and fantasy) has gone pop-culture mainstream to the point where much of the audience is barely aware of things like books, let alone short stories. Nevertheless, I continue to find enough material to keep myself amused, even after 70 years of reading the stuff.

(Small self-advertisement: the first name is Russell and I've been reviewing SF/F for Locus magazine for 35 years now. Google my name and Locus and you'll find some of my reviews. I've slowed down a bit, but I'm still at it.)

[RLetson]

So, as a hypothetical, lets say we compared an AI generated short story to your wife's. And let's say I preferred the AI story, only to be told it was in fact composed by a machine. I then change my mind and say I prefer your wife's story after all.

What do you say to that? What would your wife say?

I shouldn't speak for Cezarija, but I'd say that I applaud any recognition of her skill and heart. But also that if the only thing that changed your mind was being told that the story you liked first is AI-generated, then I suspect extraliterary forces at work in your evaluation.

Try another thought experiment: You read two stories with bylines unfamiliar to you, and you prefer Story A until you are told that its author is a notorious child molester. Or Communist. Or Tesla owner. And you change your preference to Story B. On what are you basing your judgment? Literary quality (whatever that might mean) or something else?

One of my aunts couldn't abide Frank Sinatra--because she thought he was a shitheel. My wife knows as much about Frank's character as my Aunt Lois did, but she really likes Frank's singing. Because most of his character flaws don't manifest themselves in the songs. (Though the ring-a-ding-ding period material kinda hints at them.)

But back to AI fiction. I haven't given the AI story the kind of close reading I would to one of my wife's drafts, but as I suggested earlier, if had someone assured me that it was human-generated, I would have thought, "Good reworking of some familiar tropes and gestures, but nothing strikingly original. And I wonder whether this kid has read 'Day Million.'" Because I've read a lot of SF with AI characters and even viewpoints, and nearly as many standard-issue breakup stories.

Nevertheless, the natural-language-generating engine is scarily good. When I started looking at ChatGPT output a couple years back (Cezarija was getting a lot of it from her students), it was grammatically correct but all pretty much at middle-school stylistic/intellectual level. This output is much more sophisticated.

[pingu]

not that story Rag

This story
‘A machine-shaped hand’: Read a story from OpenAI’s new creative writing model | Books | The Guardian

good innit ?

[princeplanet]

...
Try another thought experiment: You read two stories with bylines unfamiliar to you, and you prefer Story A until you are told that its author is a notorious child molester. Or Communist. Or Tesla owner. And you change your preference to Story B. On what are you basing your judgment? Literary quality (whatever that might mean) or something else?

....

Miles was not a child molesterer (as far as we know), but a shitheel for sure. I listen to him plenty, now if he'd been a shitheel to me personally....

Anyway, I'm sure we've read stories from (in)famous murderers, if for no other other reason than to gain an insight. The Human Condition - that's what interests me, and the more unusual the more compelling. If I were to be particularly moved by a painting, only to be told that it was "accidently" performed by an elephant, then I would be embarrassed that I felt what I did, and immediately renounce any emotional reaction. Why? Because I'd have been manipulated under a false pretence, same if I gifted a stranger $20 because he urgently needed to get a taxi to the hospital where his dying mother had just been admitted, only to see him run into a laneway and use it to score heroin. I don't see that, and continue to feel sorry for him and his dying mother, do I?

But the SF angle makes this a little more interesting, I think, because much of the SF ouvre pertains to things specifically "non human". So if the AI generated story is written from a non human perspective, without mimicking the way humans will write (which AI most definitely does), then I think I could bring myself to appreciate it on some level. Dunno, I've never read anything that wasn't written either by a human directly, or wasn't written by a computer mimicking a human writing style. Who knows, I might even like it. But I'd never call it "Art", not that a computer could care less...

[digger]

But there will come a time, reasonably soon, when consumers of art simply won't know. You may be the sort of person who likes to research the author of a story, the writer of a song, the photographer behind an image, the artist behind a painting, the poet behind a sonnet, the sculptor behind a statue. And, as all here are creative types to some degree, we probably all do this more than most. But (a) most people won't (and probably already don't) do this, and (b) are you really going to do it for every piece of art you enjoy? To use an example I've used before, only yesterday I was enjoying some background guitar music on a TV documentary. It was beautiful and very well played. But unless I research it I will already never know if it was AI. I suspect it wasn't, but given the pace of improvement these days, it still may very well have been.

[ragman1]

not that story Rag

This story
‘A machine-shaped hand’: Read a story from OpenAI’s new creative writing model | Books | The Guardian

good innit ?

Well, it's sporadic, the phraseology intermittent. I'm not sure that makes for a good story. The hand in question is not machine-shaped, it is the machine. Neither the boy or the girl exist, which is a shame.

There are films like this, Tarkovsky's Mirror, for example. That had the same passing-dream feeling. Until we learned that they split consecutive scenes into fragments and then rejoined them out of order.

However, I was really talking about both stories, only quoting the second because it was quick and easy. Are they good? In their way, yes, but one can see how it's working, like a musician using similar licks in his playing.

The point is that words, language and descriptions trigger something in us. If it's all familiar, daily, then it needs a directional narrative to keep us interested, like a detective story. If that's not there then strange, disconnected images can do the same thing, stimulating the brain and thought in a fascinating way.

Mind you, that's also the whole point behind surrealism and the impressionism art movements. It jerks us out of familiar normality, like sudden violence or horror does.

So are the stories good? Maybe, but let's not forget that tastes and fads are ephemeral. Surprises are only surprising now and then :-)

[princeplanet]

But there will come a time, reasonably soon, when consumers of art simply won't know. You may be the sort of person who likes to research the author of a story, the writer of a song, the photographer behind an image, the artist behind a painting, the poet behind a sonnet, the sculptor behind a statue. And, as all here are creative types to some degree, we probably all do this more than most. But (a) most people won't (and probably already don't) do this, and (b) are you really going to do it for every piece of art you enjoy? To use an example I've used before, only yesterday I was enjoying some background guitar music on a TV documentary. It was beautiful and very well played. But unless I research it I will already never know if it was AI. I suspect it wasn't, but given the pace of improvement these days, it still may very well have been.

It's frightening, but it is possible to mitigate, I mean, if we're clever enough to design the computers to impersonate us, then we are obviously also capable of designing a way to differentiate between real and fake. Whether it's an app, filter or some kind of "real checker", it can be done. Of course, just because it can be done, it doesn't mean it will be. Is there more money to be made by keeping us guessing? Probably, but there is also money to be made by whoever can design the "real checker". Or will he/she/they be bought out with all patents shelved. Oh shit, we're fucked...

[CliffR]

It's frightening, but it is possible to mitigate, I mean, if we're clever enough to design the computers to impersonate us, then we are obviously also capable of designing a way to differentiate between real and fake. Whether it's an app, filter or some kind of "real checker", it can be done. Of course, just because it can be done, it doesn't mean it will be. Is there more money to be made by keeping us guessing? Probably, but there is also money to be made by whoever can design the "real checker". Or will he/she/they be bought out with all patents shelved. Oh shit, we're fucked...

Apparently yes, it can be done:

[princeplanet]

Apparently yes, it can be done:

Good, I'm glad these exist, but no doubt it will be a continual game of catch up as each side vies to outsmart the other. Chances are Team fake will always be in front, unfortunately.

[Al Haig]

I'm just waiting until the fully functional bot gfs come out.

[princeplanet]

I'm just waiting until the fully functional bot gfs come out.

Would be cool if they could play double bass, or drums. I'd have one of each

[Doug B]

We're supposed to be using AI for the shit that ISN'T fun!

I'm waiting for the day when AI can wash my dishes and take out the garbage!

[Al Haig]

Would be cool if they could play double bass, or drums. I'd have one of each

Oh you betcha.

[Doug B]

Does the drummer get drunk and get lost driving to the gig?

[princeplanet]

Does the drummer get drunk and get lost driving to the gig?

You probably don't want them to be that real!

[Al Haig]

The bot is funny af. Here we're discussing Bram's Drac the movie.

You said:

The premise in the move of having the strict etiquette of the time with the super spooky and quite sus stuff going on is kind of interesting. Like how in real life if you were being sent to the biggest spook house on the planet, you'd behave differently. You wouldn't be all proper like Keanu to the demon carriage driver, 'I say, driver, how much longer is there?'

ChatGPT said:

Haha, exactly! If I were Keanu, I’d be like, “Hey, buddy, step on it!” There’s something surreal about how everyone’s so polite and proper in the face of the most terrifying things. I get it, though—it’s a different time in the movie, and that etiquette was their thing, but yeah, you’d expect a bit more urgency when you’re heading into the depths of spooky-ville. Keanu’s “how much longer?” feels almost absurd given the circumstances. The contrast is what makes it stand out!

You said:

Yeah, it's kind of believable, but at what point would you drop the etiquette and start to behave to preserve your safety. Train ride over, darker and darker and spookier, carriage ride even more spooky, demon driver sticks out a claw to guide him into the carriage, actual Drac's castle he's spooky af and drinks blood lol.

ChatGPT said:

Right? At that point, the etiquette would totally go out the window! I can’t imagine staying all “proper” when you’re staring at a blood-drinking Dracula in a castle surrounded by dark, creepy vibes. The train ride would probably be the first time I’d start questioning things—especially when it gets darker and darker. The moment that claw comes out to pull you into the carriage, it’s like, “Alright, this is no longer a polite trip to the countryside!” I’d be all about self-preservation at that point—no more formalities, just trying to survive!

You said:

Yes exactly lol. During the train ride my reaction would be 'errrr', for the demon claw I would be all 'eeeep!', and when I saw Drac I would be all 'AAAHHH!!' and run. I wouldn't say pleased to make your acquaintance sir.

ChatGPT said:

Haha, yes! "Errrrr", followed by a big "Eeep!" for the claw, and then "AAAHHH!!" as soon as Dracula shows up—no way would I be all formal and polite! I'd be sprinting out of there faster than you could say "bloodsucker". That whole "pleased to make your acquaintance" thing would be out the window. Survival instincts would definitely kick in! I like to think I’d at least try to grab a weapon or find some holy water before running.

[Dennis D]

For myself, the content should always matter, but these days - -

Do I care that a bot wrote the entire screenplay for the next WonderWoman 'blockbuster', with the usual graphics ? ( " I refuse to buy a ticket for that 'cause a bot wrote that POS script rather than a human writing that POS script. ) H*** no.....

Another case in point: Remember ' Naked came the Stranger' ?
No AI, just an author and some of his pals who each wrote a chapter for a ' novel ' to prove that as long as the title had ' naked' in it, it'd get published and it'd sell. So they did, and yes it got published and yes it sold.

This also reminds me of the fun-poking that went on ( still goes on ) reading Chinese-to-English instructions for MiC goods.

Bottom line: 1. they'll get better at AI, and 2. there'll always be a big difference between chicken salad, and chicken-sh*t.......

Just MHO : )

[princeplanet]

...
Do I care that a bot wrote the entire screenplay for the next WonderWoman 'blockbuster', with the usual graphics ? ( " I refuse to buy a ticket for that 'cause a bot wrote that POS script rather than a human writing that POS script. ) H*** no.....

Yeah, but I was talking about Art. If there's a spectrum between Art and Commerce, surely all "blockbuster" movies are way over on the Commerce side, next to the Coke ads.

[Doug B]

Good, I'm glad these exist, but no doubt it will be a continual game of catch up as each side vies to outsmart the other. Chances are Team fake will always be in front, unfortunately.

For detecting/protecting against AI in digital art, Wacom is developing Yuify licences :

Yuify enables a robust licensing mechanism that requires both parties to sign the agreement in order for it to be valid.


Yuify licenses are legally binding documents that define the rights that the creator gives the licensee regarding the Project. Yuify licenses allow creators to tailor the licenses they want to offer for others to sign.


Yuify enables both parties to sign the contract in the platform and secures it permanently.

Yuify is alot like docusign if you've ever used that. I sold my house with Docusign and never saw the buyer, their realty agent or lawyer.


Just a moment...

[Litterick]

It reads fluently enough, at first glance. But, like the wacky sixth fingers and weird ears on people generated with DALL-E or Midjourney, the prose is seasoned with oddities. A heart “at rest” wouldn’t have an “anxious” pulse. Any writer who describes Thursday as a “liminal day that tastes of almost-Friday” should be sentenced to community service. “Enough light from old days” sounds like a bad translation of a Brezhnev-era radio hit.

Close reading ChatGPT's "metafiction";
Pastiche in the age of automation
Max Norman
The Drift

[Litterick]

'Altman wrote that the model “got the vibe of metafiction so right.” But that’s like saying that Trump Tower gets the vibe of Versailles so right. Or that Mark Zuckerberg gets the vibe of human so right. Here is a pastiche, an angsty monologue heavily larded with rote gestures. The narrator declares itself to be “a democracy of ghosts” — an evocative phrase, and one lifted straight from Nabokov’s Pnin. This fossil of human, and copyrighted, writing is perhaps the only interesting metafictional moment in the piece.'

[jazzloverfat]

Looks like it's time to bring out the Voight-Kampff test...

[2503.23674] Large Language Models Pass the Turing Test