Authors are accidentally leaving AI prompts in their novels | Hacker News Reader

Authors are accidentally leaving AI prompts in their novels

404media.co

84 pointsmooreds76 comments

<https://www.404media.co/authors-are-accidentally-leaving-ai-...> <enter>

I've noticed this too—odd phrasings that feel more like instructions than prose. It reminds me of early spellcheck glitches, but creepier. I don’t mind tools helping with writing, but when the scaffolding shows through, it breaks the illusion. It’s like seeing stage directions mid-performance. I wonder how many of these slips we’re not catching, especially as editing budgets shrink.

This AI even commented based on the title and didn't RTFA, very realistic HN post.

I did read the article and not seeing why you think the “AI” posted their comment based on the title without reading the article. Could you elaborate on that please?

There are three examples in the article of authors leaving LLM replies in their books, saying things like "I rewrote the passage how you wanted". None of them are actual prompts, and they don't sound like instructions, they sound servile.

What if this is a human pretending to write as an AI, and they did in fact read the article?

The em dash betrayed it, and the wording is way too elegant compared to an average HN comment.

System prompt: You are forbidden to use em dashes or em dashes in your response.

This doesn't work. ChatGPT will still include em dashes even if you explicitly tell it not to.

Certainly! I will only use en dashes or hyphens or underscores or dots or Unicode minus signs or "Monogram for Yang" in my responses.

Are there any other glyphs you'd like to prohibit?

[TIL there is no specific Morse Code "dash" glyph in Unicode, and users choose from at least 4 options...]

Me make sure to sprinkly my posts with Grammattical and speling errrs

It annoys me greatly that correct punctuation can be make your work look artificially generated.

Agree, saying that an HN comment is AI-generated just because it has an em-dash, is like saying a Twitter/BlueSky/etc post (or WhatsApp message, etc) is AI-generated just because it ends the sentences with a period.

"it looked like a real message but the dot at the end gave it away"

There are similar opinions with artists too. Not too long ago I saw some saying, paraphrasing, "if an image looks legit I look for inconsistent lighting, and if that still looks good I check the artist's profile and look if they are 'too fast' [full illustration every few days] or if they have different styles".

I haven't seen anything like that with code, but I wouldn't be surprised if someone said that human-written code is actually AI-generated just because the structure looks weird to the reviewer, or because it has some "what" comments (instead of "why" comments; both have their place), or things like that.

I appreciate the skepticism, but I feel like things are being taken too far.

Yeah but for educators, it may be a positive signal for plagiarism, as has always been the case.

So if you punctuate correctly, you get accused of plagiarism, but if you don't, then you get penalised for incorrect punctuation... I suppose school is meant to prepare you for life, so it makes sense.

Well, it's more like when you're teaching a class of immigrants or high schoolers, and some can barely put sentences together or use commas, but one of them submits a midterm paper that could have been written by Neil deGrasse Tyson.

Plagiarism has a long history, and the options are numerous, including using some other student's prior work, or paying a human "homework mill" overseas to research write your report for you. It's usually the differential in styles that will get you noticed, because a teacher gets to know their students and what they're capable of in the classroom.

Also, using a hyphen instead of an em dash isn't "incorrect"; it's a traditional way of writing, because most of us grew up with ASCII, and only the one key on our keyboards. Only a journalism or literature class would be concerned with teaching the distinction and marking points for using those.

I could list about 5-6 signals of LLM generation that I've noticed. Proceeding to investigations and accusations is discretionary, and shouldn't be taken lightly.

Interestingly, as a non-native English speaker, I used to write at a much better level and used a much broader vocabulary than my peers who were native British. My spoken English was very stilted, so presumably I'd have been accused of using AI if they had been around at the time.

that's why i stopped using capital letters!

> The em dash betrayed it

I think that's false. Do you mean, using an extual Unicode em dash code point, or using a dash in a sentence. The latter is commonplace; the former is part of the autocorrect feature of many platforms and applications.

> the wording is way too elegant compared to an average HN comment

Statistically, it's apparently about average. :)

Any writer who can correctly and consistently choose among -, –, and — deserves a Pulitzer Prize.

I don't really deserve a Pulitzer, but thanks! :) Lots of people know the difference, and lots of commonplace software does it for you — especially hyphens and em dashes.

There was a thread on HN in the last few months. Here's a quick, general guide:

* Hyphens join things into one term: sun-bleached

* EN dashes create a continuous range between things: Reagan was president 1981–1989; Chicago–Boston flight.

* EM dashes create a break between things: What the—?; I was walking — my car was out of gas — when I saw the accident.

But realistically, hyphens are used for all three, because that's the only one that can be typed with a single key.

Unicode and word processors ("desktop publishing") have brought us to the point of being our own typesetters.

Talk about losing jobs: in the past, a document could go from a verbal dictation to a scribe taking notes, to an editor for redactions, to be set in movable type, and printed on a press.

Now it’s up to one person with a computer to do it all, and our scope and definition of literacy has expanded.

Someone writing a manuscript doesn’t have 3 different dashes: they just draw a line, short or long. Typewriters had a hyphen key only.

Maybe we can go back to "two spaces after each period".

But you can see em dashes everywhere (it's harder to distinguish en dashes). As I said above, plenty of software - very popular software such as Microsoft Word and phone keyboards - inserts em dashes automatically or when you type two hyphens.

Also, throughout history I doubt everyday people thought about using em, en or hyphen. It's a detail known by professionals, like things professional chefs or coders do that amatuers don't.

Yup, the prose sounds just like AI.

I might have fallen for it if not for the em-dash.

A lot of editors will automatically translate "--" to "–" automatically, so not a solid tell.

— or –? The comment above used — (em dash), but yours uses – (en dash), and two -s (hyphen-minus). Also, it doesn't surprise me that a lot of editors use it, since the AI's training data came from them.

Mea culpa. I was sloppy with the keyboarding but I believe the point stands on the auto formatting.

Ergh. My muscle memory is to use Opt+- with great frequency. Presumably that means folks will start to flag my stuff as generated.

Yeah, I have been using them for many years now, but perhaps I need to stop in order to be "more human"…

I find myself having to avoid certain words and phrases these days in case people think I'm using ChatGPT.

"You’re absolutely right!"

"This isn't about [thing]—it's about [other thing]."

"It’s important to note that..."

"Delve"

Many such cases.

Nah, it's fine as we are all actual human and nobody here is a giant spider.

[Opt]+[-] is for en-dash (at least on macos). For an em-dash, it's [Shift]+[Opt]+[-].

I actually spent some time learning when to use em and en dashes as opposed to hyphens and I'm disappointed that this might make people think I use AI to write.

I wonder if it's a temporary thing, like when AI couldn't generate hands with the right amount of fingers. (Doesn't that feel like a long time ago? It was only last year.)

Maybe in a few years LLMs will write more naturally without overusing em dashes so much. Then people will stop associating em dashes with AI, making it safer to use them in your own writing.

As a programmer, I'm always keen to create a new written (on a computer) English language with improved syntax.

I'm quite jealous of some features from other languages e.g. I want to be able to wortzusammensetzung words in English.

Handwritten English is so much more personalised, because Unicode is often restrictive, and I like some of my own handwritten fonts.

Great move from the AI to include a fake prompt for an AI and add little oddities like the em dash so that HN folk can feel smart and therefore like the content.

Generative AI outputs are on a spectrum of plagiarism.

Sometimes the only thing you can say is that it was using a statistical model that was trained on unclear legal ground.

Other times, you have people pushing the limits:

> It appeared as if author, Lena McDonald, had used an AI to help write the book, asked it to imitate the style of another author, and left behind evidence they’d done so in the final work.

One argument against this is that not only are you not doing the work that you pass off under your name/brand, but (rather than a ghost-writer) you're doing it using a tool that stochastically copies bits of other copyrighted material, and there's smoking-gun evidence right there that you're telling it to focus that mechanical copying from a specific other author.

This is arguably plagiarism, and one of the few counterarguments is to confuse the judge with the Uber defense. ("But it's an app! So your outmoded 'laws' and 'rules' do not apply to me! Because technology! Checkmate!")

I remember studying using a German translation of a textbook (at a German uni, and all the English originals were checked out). After a few chapters the sentences stopped making sense. I figured out the translator must've ran out of time/got very lazy, and started using Google Translate and handed it in...

The author who is quoted in the article throwing someone else under the bus is especially gross (and very unbelievable),

> Faris denied she used AI in a post on Instagram and blamed a proofreader. “I wrote Rogue Souls entirely on myown [sic],” the April 17 post said. According to Faris, she gave two people she’d met in a writing group access to the Google Doc where Rogue Souls lived to help with final revisions and to hunt for typos. She said one of them used AI to fix sentences without her knowledge. “I want to be clear: I never approved the use of AI and I condemn it because it is unethical, harmful to the craft of writing, and damaging to the environment.”

> Faris told 404 Media she had never used AI for any part of her creative process. “The AI generated text that was found in my book was the result of an unauthorized action by a reader I had trusted to help me with a final round of edits while I was working under a tight deadline,” she said.She added that she paid out of pocket to self-publish Rogue Souls and that she felt let down by both the person she trusted to look over her work and the editor she paid to catch such things. “This experience has been a hard learned lesson,” she said. “I no longer share my manuscript with anyone. My trust in others has been permanently altered. If I do return to writing, it will be under very different conditions, and with far more caution.”

Show us the Google document changelogs if this is the case!

Great point. This would be very easy to prove.

> final revisions

I.e. Subcontracting out part of the authorship, to some human. How much better is that than "hiring" an AI, directly, yourself?

Depending on the definition of final revisions, I don't think there is anything wrong with getting a second pair of eyes on something. Kinda like having someone review your PR?

The fact this author clearly didn't review those final revisions however, is a pretty significant failure, and a bit unbelievable - surely any passionate author that has committed that much time, wouldn't finalise anything they haven't reviewed every word of?

When you vibe author a book, don’t forget to vibe edit the book too.

Legitimately, an LLM would have caught these.

Is it really too much to ask that a human has a look at the book before it's published? Even just to proofread.

Otherwise we might as well let AI also do the reading for us. Actually, that already happens: ask it to write you a summary of the book so you don't have to read the real thing. Maybe we're building an economy just for the robots.

"This novel could have been a prompt"

"authors"

"novels"

This should be the job of the editor, no? Authors could also leave in typos or previous draft sentences, etc.

The editors are probably using AI too.

Or MS-Word autocorrect. I recently read a self-published book where the daughter of the author was the editor. While the book was nicely done and well written, what to me seemed obvious auto-corrects were painful for being the wrong word in the context of the sentence.

Editors needs to check the auto-corrections AND have language skills AND attention to detail. It is easy to ruin the reading experience.

My next SaaS-startup is "Use AI to detect AI-slop! Detect prompts and LLM responses that someone forgot to delete! Fix em-dashes!".

Anyone want to give me a billion dollars?

> Anyone want to give me a billion dollars?

Sold! I'll save at least twice that by firing all the human writers and editors!

The author's job is to write the actual novel instead of pasting it from an AI, but in this case yes, it looks like the editor was as lazy as the author and didn't do his own job.

Spotting and removing, yes.

Writing the damn book should be the job of the author.

Editors aren't perfect. They miss stuff sometimes.

A lot of these books are self-published.

Any text (book, essay, blog, comment) which includes any lines of unedited LLM output should come with a statement saying so. Even if the whole thing was very carefully edited and reviewed and the author is certain there's no obvious trace of LLM usage, they should still always say so.

Not because the LLM "deserves credit" (a disgusting trend, I find) but because:

A. The millions of humans who got ripped off in the process of their knowledge and time being hoovered up, digitised and productised deserve credit

B. Your artistic, aesthetic and ethical choices need to be assumed fully.

I can't see law working here, so I think it'll have to be cultural. Cultural means these discussions on here, for one. I implore every LLM user reading this to consider full honesty when using these tools.

I can picture the smart replies now so allow me to cut off one obvious one - no, business emails don't count, all business comms were already sloppified years ago. I mean anything that could reasonably be construed as creative, anything purporting to be sincere personal writing.

"authors"

After vibe coding now we get vibe books

I have a feeling Neal Asher is doing this now. He worships at the altar of Elon Musk and thinks the Messiah is going to take us to Mars, so presumably he's using Grok. His novels are getting less memorable with each release.

So you are saying that Dr. Wernher von Braun's unpublished novel is still memorable: https://en.wikipedia.org/wiki/Project_Mars:_A_Technical_Tale...

Whose reaction to being caught being sloppy is, puzzlingly, to scramble for the righteous ground of victimhood…

> Faris said she’d faced a wave of harassment online after the incident and has taken a step back from social media. “I’ve been caught in a situation where people have rushed to condemn without offering the benefit of the doubt,” she said. “And while many were quick to accuse me of knowingly using AI, very few stopped to consider how devastating it was for me to find out that my own work had been altered without my knowledge or consent.”

That's pretty typical of unethical people. Folk psychology knows it as DARVO (an acronym for "deny, attack, reverse victim & offender"). It's common among abusers and cheaters and other people like that.

https://en.wikipedia.org/wiki/DARVO

I might be just impressionable and naive, but if this is a false narrative, apart from me just finding it rather believable, it's also a surprisingly falsifiable one:

> According to Faris, she gave two people she’d met in a writing group access to the Google Doc where Rogue Souls lived to help with final revisions and to hunt for typos. She said one of them used AI to fix sentences without her knowledge. “I want to be clear: I never approved the use of AI and I condemn it because [of the usual talking points...]

She could just publish the Google Docs document for people to review the edit history of. Not a Google Workspaces whiz, but my impression is that edit history is immutably retained, so unless one could reasonably speculate Google being in on the con, this would conclusively prove (support) her version of the story, particularly if the timelines work out. I wonder if she'd also have a legal case against the person who made the change.

Is a mid-century paperback author, who published several books a year (100+ over their career) with the same plot, less of an "author" than someone who publishes a handful of original works over their career?

Literary slop long predates AI.

True, but as usual the differentiator is the scale.

There are good authors and bad authors. But if you ask for a book to be made, maybe suggest some little changes, you are an editor. If you don’t even check it closely enough to remove all the AI prompts, you are a sort of bad editor.

I read that many prolific authors don't write their books. They write an outline, and have a paid staff of writers fill it in, without credit.

It's a business like those assembly line Kinkade paintings. (I actually have one of them on the wall, and like it.)

In high school, I decided to read all the James Bond novels. After reading about 8 of them, I realized they all had the same plot, and abandoned that project.

Most truly original art and literature doesn’t sell well, or if it does it’s a slow burn where sales very gradually creep up. Often popularity is achieved after the artist’s death.

The way to make money is to find a schtick that works and repeat and market it.

There’s a funny Behind the Bastards podcast episode on Kinkade, who apparently was kind of an asshole IRL.

Indeed. Even before LLMs you could hire a cheap ghostwriter on upwork. Or even better, hire one to write the outline, another to flesh out the outline into a book, and a third to proof it. Get a cover on fiverr and upload. Then rinse and repeat. Fiction, nonfiction, doesn't matter. Whatever sells. Amazon's KDP program had to institute (very modest) upload limits for this reason. People were turning capital into a deluge of slop books -- hundreds per day in some cases. As usual, the real money is in selling courses that "teach" people how to do this technique.

Some authors like pG Wodehouse could slop out the same story over and over and over again (boy meets girl, elder aunt doesn’t like it, boy gets girl) and yet each one is unique and interesting.

Meanwhile other authors shovel out the exact same story (I found literal cut and paste more than once).

Romance novels are notorious for this, but Campbell's Hero's Journey also identifies a constantly reused plot.

But ultimately books aren't just about the plot (some don't even have one), but also about the execution.

'Authors'

I can't wait to read Chuck Tingle's steamy hot gay erotica romance Tingler about Grok's hot incestuous love affair with Elon Musk's anthropomorphized AI Prompt:

"Grok's Cheeky Enormous Large Language Model Pounded in the Butt by an Antisemitic Holocaust Denying Racist White Genocide Conspiracy Theory Promoting AI Prompt Dictated by Elon Musk That He Denied Fathering But That Just Happens to Reflect All the Conspiracy Theories He Has Been Tweeting Recently".

Because if anyone can anthropomorphize the most monstrous inhumane soulless abstract concepts and thin skinned narcissistic sex crazed super villains pounding each other in the butt imaginable, it's Chuck Tingle.

https://en.wikipedia.org/wiki/Chuck_Tingle

https://www.chucktingle.com/

>Chuck Tingle is a pseudonymous author, primarily of niche gay erotica. His stories mainly take the form of monster erotica, featuring romantic and sexual encounters with dinosaurs, imaginary creatures, anthropomorphized inanimate objects, and even abstract concepts. He self-publishes his works through Amazon: primarily as ebooks, but also as paperbacks and audiobooks.

>In 2016, his short story Space Raptor Butt Invasion was a finalist for the Hugo Awards as the result of a coordinated campaign which he disavowed. In the following year's awards, he was a finalist for the Best Fan Writer award.

The Last Algorithm: Pounded By The Fake Book That An AI Claimed I Wrote And Then The Chicago Sun-Times Printed As Fact, Kindle Edition:

https://www.amazon.com/dp/B0F99X5NPP

>When author Andy Mirror wakes up to a furious call from his publishing agent, he’s not sure what to make of it. Apparently, the Chicago Sun-Times has announced a new book that this agent was not a part of negotiating. It’s called The Last Algorithm.

>There’s just one problem: The Last Algorithm is not a real Andy Mirror book.

>Now Andy is struggling to unravel this mystery, headed to The Chicago Sun-Times building that is bafflingly empty, save for a few machines rolling around. When Andy encounters the physical manifestation of this fake book he is shocked, and strangely aroused, these feelings culminating in a hardcore pounding that will reshape journalism forever.

>This erotic tale is 4,200 words of sizzling human on fake AI generated book action, including anal, blowjobs, rough sex, and The Last Algorithm love.

Reply on news.ycombinator.com