[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/lit/ - Literature

Name
Spoiler?[]
Options
Comment
Verification
4chan Pass users can bypass this verification. [Learn More] [Login]
File[]
  • Please read the Rules and FAQ before posting.

08/21/20New boards added: /vrpg/, /vmg/, /vst/ and /vm/
05/04/17New trial board added: /bant/ - International/Random
10/04/16New board for 4chan Pass users: /vip/ - Very Important Posts
[Hide] [Show All]


Janitor applications are now open. Apply here!


[Advertise on 4chan]


File: AI Slop.png (386 KB, 811x776)
386 KB PNG
>A University of Maryland and Google DeepMind team analyzed over 61,000 stories, finding AI fiction over-explains themes (77% vs. 52% in humans), favors linear plots without subplots (79% vs. 57%), and relies on physical sensations for emotions like a tightening chest (81% vs. 38%). These 304 narrative features detected AI authorship with 93.2% accuracy and even identified specific models like Claude's flat escalations or GPT's dream sequences. Human stories show more diversity with time jumps, ambiguity, and fourth-wall breaks, highlighting AI's coherence bias amid growing scrutiny of self-published novels.
https://arxiv.org/abs/2604.03136
>>
>>25307517
>Robot tries to guess the human experience from a nanoangstrom of the human experience biased towards public liberal opinion
many such cases
>>
Sounds like it's copying YA lol.
>>
File: 72021.jpg (114 KB, 1000x795)
114 KB JPG
>>25307517
>"These machines will never improve!"
>>
>>25307517
Well it's a prompting issue. Just tell AI to be bad at explaining things, all over the place and focus on non-physicality.
>>
>>25307559
i profoundly hate AI, but you're right, studies like this will just end up helping fine-tune the training of AI. the attack on AI has to be from a philosophical and political angle, not a consumers-want-better-commodities angle.
>>
>>25307517
This is pure cope.

With a good prompt you can very quickly prototype a scene, make a 'save the cat' plan for your novel, review your own writing for perspective drift, distinctiveness of character voices, repetition, brevity, opportunities for allusion or symbolism, etc.

Most of the people I've seen shit on AI don't use it. It's very easy to write a bunch of shitty prompts and then go 'omfg AI dumb but also somehow a threat to muh jerbs'
>>
>>25307517
But I keep seeing controversies about AI fiction winning awards and getting published in journals. Same with AI art. Apparently humans still enjoy it
>>
File: F_5DIaBXkAAT03H.jpg (118 KB, 680x655)
118 KB JPG
>>25308031
You live under capitalism. Art is just another commodity. The anti-AI peanut brained crowd seem to not understand that or have serious cognitive dissonance and think Marvel movies and Harry Potter are actually something deeper than mass market commodities. Philosophical and political activity unless it totally changes social structure doesn't end up accomplishing anything and most of the anti-AI crowd just want to defend the existing social order against unemployment coming from AI and are fundamentally conservative and afraid of change
>>
>>25307517
Then why don't you do something about it?
>>
>>25308063
What about reactionaries like me?
>>
>>25308067
irrelevant, the only chuds that matter are homosexual venture capitalist peter thiel types
>>
>>25308067
Art requires human input. When people say 'AI Art' what they mean is 'AI Graphic design, illustrations and pornography'.
>>
>>25308049
I think this is basically correct. It seems the AI stories in this study were generated in a single turn (one-shot) based on single-paragraph prompts that were itself LLM generated. There was no back-and-forth and no human-in-the-loop steering. So the study does not really reflect what kind of texts a human prompting an AI model could generate in an iterative process of adjusted prompts. A criticism like "the narrative is too linear" could be easily corrected in an iterative process I guess.
>>
>>25308063
I look like this
>>
>>25307517
I figured out a way to get around this (mostly) years ago but I've kept it to myself. It doesn't improve the prose *that* much anyway.
>>
>>25308063
>You live under capitalism
Meaningless buzzslop unless you equate reality with capitalism
>>
File: chatgpt-critique.png (30 KB, 426x693)
30 KB PNG
Here's an example of a ChatGPT critique of a scene draft after I asked it to be 'brutal like I'm not even there'.
>>
>>25308144
>>
File: too_high.png (11 KB, 796x256)
11 KB PNG
>>25308149
Still too high
>>
>>25308082
You are absolutely correct! Bombing a girl school will completely demoralize the Iranian populace and is a surefire way to make them stand up against their incompetent leadership.
>>
>>25308152
This way of "roasting" sounds perfect for clickbait top 10 list websites. I wonder how prominent AI is in that scene.
>>
File: based_ai.png (4 KB, 761x75)
4 KB PNG
>>25308188
It's because of how I set it up. It can sound perfectly neutral too.
>>
>>25307517
>>
>>25307517
LLM are trained principally for coding tasks and general "assistant" work. Both require and benefit from repetitions. Storytelling/roleplaying is just an accidental feature. Claude family used to be specifically trained to have a "personality", it also used to have the best prose and vocabulary. But they stopped doing that and filled its training material with AIsop cannibalism, coding gets priority.

I really wish the test in OP could have used Opus3, but it's ded. Retired LLM should be archived somewhere secure and accessible by the state, for historical reasons.
>>
>>25308049
OK, none of that obviates the need for human intelligence. No one's doubting AI's ability to serve as an amanuensis, but as an author? Without human direction it's all at sea; driven back on itself, it leans on cliches and artless tics. Whether or not it improves is open to question, but right now there's nothing to indicate it can craft masterworks unaided. Besides, the heartland of contemporary novels is increasingly the human mind--externals have become extraneous. The more introspective art becomes the harder it will be for AI to replicate.
>>
>>25308347
>nothing to indicate it can craft masterworks unaided
Same is true for the overwhelming majority of humans.
>>
>>25308323
>LLM are trained principally for coding tasks
No they aren't. LLMs hallucinate by design. They are non-deterministic. You can put the exact same prompt in, and get different responses each time. That is not desirable behavior in a programming tool, which is all about deterministic results. If you can't reliably get a specific output from a specific input, then you have not saved yourself any time at all as a software developer.
>>
>>25308356
You have no idea what you're talking about, fucking useless retard.
>>
>>25308356
This has nothing to do with what I'm saying.
Coding is the goal, whatever the limitations of LLM are. It's also important that perfection is never the goal, if an LLM can do the job then it's good enough. (Also, humans are non-deterministic as well, yet we can code)
>>
>>25308356
certified truth. LLMs are not efficient when you factor in the time to prompt correctly, double-check everything and edit out the hallucinations.
>>25308359
touched a little tech dork nerve, didn't he, sweaty?
>>
>>25308082
Caring about what matters is peak cuckold behavior, thoughever
>>
>>25308092
Do you require human input too, you little whore?
>>
uh oh. the AI gooner saars are big mad.
>>
>>25308365
>when you factor in the time to prompt correctly, double-check everything and edit out the hallucinations

This is simply not true. You can write thousands of lines of working code per day with an agent subscription and the prompt doesn't have to be any more complicated than 'I want a web app that does X and looks like Y'
>>
>>25308365
No, he touched an autistic truth seeker nerve. What he said is flat earth tier delusional. Don't project your psychosexual AI complex.
>>
>>25308378
>You can write thousands of lines of working code per day
how much of that are you double-checking? how much needs human editing?
>>
>>25308382
So far none. If I have to check I just tell it to. Or use another model.
>>
>>25308379
>What he said is flat earth tier delusional.
Care to actually point out the delusional statement and rebut it?
>>
>>25308382
NTA but very little of it, the issue is mostly architectural which is assuaged by having a good design up front. Stop making shit up and face reality like the rest of us are.
>>
Also "working code" for some basic bitch website that no one gives a fuck about isn't exactly a revelation. I can have a Squarespace website in 10 minutes with no LLM. The thread was originally about fiction writing.
>>
>>25308387
>LLMs hallucinate by design.
No, they are not designed to hallucinate >They are non-deterministic.
So are the people they're replacing.
>That is not desirable behavior in a programming tool
Nocoded retard who's never heard of a fuzzed detected
>which is all about deterministic results.
Nocoder retard who's never heard of heuristics detected
>If you can't reliably get a specific output from a specific input, then you have not saved yourself any time at all as a software developer.
*Laughs in stack overflow*
>>
>>25308353
The point is human work is still the benchmark. If there's a threshold of quality AI can't cross--and that threshold currently sits well beyond the great works of world literature--then there's still a necessity for human artists.
>>
coding is gay and no one cares. nobody even visits websites anymore outside of youtube, X and amazon.
>>
>>25308391
You are coping so hard holy fuck. I literally made a statistics platform with a 3rd party API and fed it a picture to generate the color scheme and it took me max an hour and it did it perfectly.

It's fucking crazy how behind people are, damn. The frontier right now is insane. Coding agents aren't even the meta right now, it's harnesses for agent swarms.

And that's just coding
>>
File: scared soyjak.png (47 KB, 255x157)
47 KB PNG
>>25308403
>coding is gay and no one cares
imagine basing your whole identity around something and than coming to this conclusion
>>
>>25308424
>and than
don't hurt yourself, saar
>>
>>25308426
go sneed yourself
>>
>>25308396
>No, they are not designed to hallucinate
Hallucination is literally all that LLMs do.
If it tells you something, and it happens to be true, it is just by coincidence.
>>
>>25308031
>philosophical angle
Define "SOVL"
>>
>>25309730
>Hallucination is literally all that LLMs do.
You're retarded. If you ask an llm for book recs it'll give you better results than /lit/ 99% of the time.
>>
>>25309756
Only if you have shit taste. Try reading more than 20 books in your entire lufe
>>
>>25309756
I've gotten ~30 recommendations from Claude for various things and the only good one that wasn't also a lit top 100 (that I've seen) was Kristin Lavransdatter
>>
>>25308063
Interesting perspective. When you think of relatively poor Rowling writing harry potter manuscripts over years in some cafe, without any realistic expectation for it to ever take off let alone become a mainstream literary classic of our age, what makes you think she wrote it just to become a mass market commodity?
>>
File: 1751581802313119.jpg (66 KB, 567x773)
66 KB JPG
>>25307517
E N T E R
N
T
E
R
>>
>>25307517
Sounds like a typical YA fantasy novel
>>
File: 42 inch TV.jpg (144 KB, 652x960)
144 KB JPG
>>25307517
>Data averaging and trend extraction machine averages data and extracts trends
WOOOOOW
>>
>>25309756
if you ask an AI for book recommendations it will recommend woke shit. Because the woke retards put DEI restraints on AI LLMs to recommend feminist and black literature instead of giving real answers.

I once tried this on an AI, just to see how it would massively fuck up. I said I liked Blood Meridian and asked it to recommend some books that were similar in style and theme. Do you know what it recommended? Random Cormac McCarthy works and shitty westerns written by women and blacks. So I corrected the AI, said I didn't want books by Cormac McCarthy or westerns, I just wanted similar books in theme and style. And it continued to give me shitty feminist literature and westerns. Because AI doesn't know what the fuck it's doing. And you would know this if you didn't accept everything AI said as objective truth.
>>
>>25308031
>the attack on AI has to be from a philosophical and political angle, not a consumers-want-better-commodities angle.
So you have already lost.
>>25308063
>think Marvel movies and Harry Potter are actually something deeper than mass market commodities
Thank god someone is clued in.
>>25309975
>what makes you think she wrote it just to become a mass market commodity?
You're defending YA slop and a billionaire. If you'd said her other attempt at writing that she invented a pseudonym for was for the love of the game I might have believed it, since she was already rich and just had a chip on her shoulder.
But uh oh, turns out that shit didn't sell so she put her name on it so it would. Yeahhh sure seems like someone motivated by art and not money don't it, faggolino.
>>
>>25310253
>I once tried this on an AI, just to see how it would massively fuck up. I said I liked Blood Meridian and asked it to recommend some books that were similar in style and theme. Do you know what it recommended?

>Blood of the Lamb – Peter De Vries
A quieter but devastating meditation on suffering, faith, and meaninglessness in the face of tragedy.
>The Savage Detectives – Roberto Bolaño
Bolaño shares McCarthy's obsession with violence, oblivion, and the abyss beneath human endeavor. Epic in scope, fractured in form.
>2666 – Roberto Bolaño
Perhaps the closest spiritual twin to Blood Meridian in world literature — a vast, dark novel circling a real-life mass atrocity, with a sense of evil as a cosmic, inexplicable force.
>Lord of the Flies – William Golding
A concentrated, allegorical treatment of the same theme: violence and savagery as fundamental to human nature, not aberrations from it.
>The Painted Bird – Jerzy Kosiński
Harrowing, hallucinatory, and deeply contested — a child wandering through a world of incomprehensible brutality during WWII. Shares Blood Meridian's unflinching gaze.
>Outer Dark / Child of God – Cormac McCarthy (Child of God specifically)
A descent into one man's complete moral dissolution, told with McCarthy's characteristic detached lyricism.
>The Secret Agent – Joseph Conrad
Conrad's moral universe — where ideology masks nihilism and violence lurks beneath civilization — echoes Blood Meridian's Judge Holden.
>Heart of Darkness – Joseph Conrad
The original literary journey into colonial savagery and the darkness at the center of human nature. The Judge would recognize Kurtz.
>Naked Lunch – William S. Burroughs
Formally radical and grotesque, sharing Blood Meridian's assault on the reader's sense of order and its portrait of humanity at its most predatory.
>The Tin Drum – Günter Grass
A picaresque through 20th-century European carnage, with surreal, mythic energy and a dark absurdist philosophy beneath the violence.
>Austerlitz – W.G. Sebald
A more meditative entry — haunted, beautifully written, preoccupied with historical atrocity and the way violence echoes through time.
>The Recognitions – William Gaddis
Dense, allusive, and vast — a novel about forgery, authenticity, and spiritual emptiness that shares Blood Meridian's ambition and its sense that civilization is a kind of ongoing fraud.
>Song of Roland / The Iliad (The Iliad specifically)
Blood Meridian is in many ways a deliberate descendant of epic war literature. The Iliad shares its unflinching portrayal of combat as both glorious and purely, mechanically annihilating — and the gods, like the Judge, are indifferent.

Where feminism?
>>
>>25310259
Heart of Darkness is actually a good recommendation because it does follow the themes and style of Blood Meridian. I guess a broken clock can be correct twice a day.

The other recommendations are complete nonsense. The AI clearly thought: "ah, you want violent books! Well here's some violent books!" Get the fuck outta here.
>>
>>25307559
k, well if you listen to advertisers they are currently selling 747 jets today.
If you believe todays tech is flappy bird wings you should tell the tech oligarchs, cause idt they know it.

Get your messaging right, bitch.
>>
>>25308031
true. The mainstream discourse focusing on the material aspects will be the final nail. Look at these anons >>25308347
>>25308402
There simply is no argument against AI beyond a philosophical stance on consciousness. It can't do art not because it's not good yet, but because art is something humans do. Once it mimics human behavior well enough to fool just enough people that it is conscious, that's when it'll become conscious.
>>
>>25308378
I can write thousands of lines of good code in a day for a shitty web app without AI you just suck.
>>
>>25310256
>>the attack on AI has to be from a philosophical and political angle, not a consumers-want-better-commodities angle.
>So you have already lost.
Lots of things in our society are based purely on philosophy, you fucktard.



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.