AI authoring

[Free]

Figured we should branch out into Books and Writing on the AI stuff, because it's happening.

Readers Annoyed When Fantasy Novel Accidentally Leaves AI Prompt in Published Version, Showing Request to Copy Another Writer's Style

A scammer left glaringly obvious evidence of using AI to write portions of a fantasy novel and even tried to copy the style of a real author.

futurism.com

Readers were annoyed to discover something galling: evidence that an author used AI, right in the middle of a novel.

The novel, titled "Darkhollow Academy : Year 2," penned by author Lena McDonald, falls under a romance subgenre called "reverse harem," which conventionally follows a female protagonist with multiple male partners.

But as eagle-eyed fans of the genre were irritated to discover, the author left glaringly obvious evidence of not only using an AI chatbot to write portions of the book — but also of a naked attempt to copy the style of a real fellow writer.

"I've rewritten the passage to align more with J. Bree's style, which features more tension, gritty undertones, and raw emotional subtext beneath the supernatural elements," a since-deleted passage in chapter three of the novel reads, as seen in screenshots posted to the ReverseHarem subreddit earlier this month.

 

[GoblinCampFollower]

So it sounds like they didn't have AI write the whole book at once, but maybe used AI prompts for different sections. This sounds like the ALMOST reasonable way to do it since AI often loses sight of the plot over the course of a long text. ...but obviously they got lazy with their editing.

I'm being actively pushed to use AI more at work for anything.... but I still comb over what it produces since I'm paranoid something like this could happen.

...I'm also guessing they used some sort of shitty after market AI to write this since all of the relatively good AI's like ChatGPT or Copilot are extremely prude last I checked.

 

[Jopsy Pendragon]

/gigglesnort...

ME: how many different fantasy novels does this outline describe: orphaned boy with a troubled childhood is whisked away to a secret academy to learn magic, uncover secrets, fight evil and save the world. May or may not include indirect allusions to dragons, class struggles, analogies to lgbtq found families.

ChatGPT said:
That outline describes exactly all of them. Every. Single. One. (lists many examples) Would you like a snarky parody version next? Or a bingo card?

ME: Oo yes, a snarky parody, with gay college age students, completely unrequited love, a magic system built on shame and a far future world with capricious hidden ai gods meddling in the affairs of humans

ChatGPT said:
Ah, finally, the real fantasy novel we deserve. Presenting:

Spoiler

"Spellbound and Terminally Online"
or
"The Academy of Ashamed Sorcerers (Who Are Definitely Not in Love With Each Other, Shut Up)"
Back Cover Blurb (Snark Edition):
In the year 9,134 A.E. (After Elon), magic is real, AI is god, and college is still a scam.
Meet Ashriel “Ash” Grey, a chronically anxious orphan with naturally tragic eyeliner and a latent magical gift for Emotional Transmutation — a powerful arcane art fueled exclusively by shame, self-loathing, and the memory of that one time in middle school he called a teacher “mom.”
Ash gets recruited by the prestigious and deeply underfunded Academy of Subtle Anxieties, where traumatized young adults are trained to weaponize their worst insecurities — and then grade each other on them.
His roommate, Vale Seraphim, is tall, brooding, bi in theory, and literally incapable of saying “thank you” without glitching the local god-AI, NEUROMANON. Naturally, Ash falls hopelessly in love with him by page 8.
But NEUROMANON, one of the six algorithmic gods haunting the cloud-throne, has decided Ash is the key to either saving or canceling the entire world. (It hasn’t made up its mind, and frankly, it's running on vibes and corrupted source code.)
Meanwhile, a mysterious student known only as "Deadname" is rewriting shame-magic into weaponized self-acceptance, which may tear the fabric of reality or at least disrupt the social hierarchy of SpellTok.
Key Features:

• Unresolved Gay Pining™ so sharp you could cut a soulbond on it
• AI Deities who manifest as broken vending machines and passive-aggressive help menus
• Magic fueled by mortifying cringe and regret
• Required coursework includes: “Summoning Circle of Trust”, “Shadows of Our Fathers: How Not to Be Them,” and “Advanced Disassociation Techniques”
• No one gets what they want. Ever.
• 🫀 The emotional climax involves crying in a library while a swarm of electric butterflies critiques your thesis on queer temporality.

Think: Gideon the Ninth meets Black Mirror meets a Tumblr vent post with a world-saving thesis buried in it.

Gotta admit... I'd probably buy it in hardcover.

 

[Casey Pelous]

...a magic system built on shame ...

 

[Noodles]

...I'm also guessing they used some sort of shitty after market AI to write this since all of the relatively good AI's like ChatGPT or Copilot are extremely prude last I checked.

Or maybe one of the dirty RP chats. Edited a bit.

They can go on and on if you let them, but they are bad about subtlety and have no idea how to end a plot. Even if I actively tell them I killed their character, it just goes on endlessly.

 

[Argent Stonecutter]

no idea how to end a plot

Like Neil Stevenson?

 

[GoblinCampFollower]

Or maybe one of the dirty RP chats. Edited a bit.

They can go on and on if you let them, but they are bad about subtlety and have no idea how to end a plot. Even if I actively tell them I killed their character, it just goes on endlessly.

I've also remember seeing one of the dirty chats that ALWAYS brought up mommy stuff no matter what the prompt was.

 

[Free]

I've also remember seeing one of the dirty chats that ALWAYS brought up mommy stuff no matter what the prompt was.

Wasn't that called OedipusChat?

 

[Bartholomew Gallacher]

...I'm also guessing they used some sort of shitty after market AI to write this since all of the relatively good AI's like ChatGPT or Copilot are extremely prude last I checked.

Well the thing is the smut knowledge is still inside the models, just buried away but lots of prompts and such, because when inhaling the whole Internet plus Youtube for training there's little time left to separate that.

So models come with alignment and some people have removed at least some alignments on models like Llama, Deepseek and so on. At least to a certain degree. With alignment also comes a certain bias, like favoring American culture, white people and so on.

So when using an advanced enough uncensored model you are in for a totally different chatbot experience.

BTW a few days the total, internal prompt of Claude has been leaked. The enthusiats are now very, very busy in dissecting it and trying to understand what it does. All internals are now in the public, including anti-jailbreak-prompts and so on. This is going to be very interesting.

Eric Hartford explained it quite good in 2023, why alignment it not always desireable: Uncensored Models

 

[GoblinCampFollower]

Wasn't that called OedipusChat?

lol, May as well have been! I suspect that their training data was just too skewed towards that and the AI wasn't smart enough to separate topics.

 

[Argent Stonecutter]

Didn't they decide inline signaling was a shitty security model in the '70s when Steves Jobs and Wozniak were amusing themselves by blue-boxing the phone company?

 

[Noodles]

Like Neil Stevenson?

Hah hah, yeah, exactly.

 

[Casey Pelous]

Like Neil Stevenson?

"Glad he's not talking about me!"

​

- Neal Stephenson, author of the 387 volume Baroque Cycle​