AI edits rejected

[theAdventurable]

I wrote a pseudoscientific (I'm not an actual scientist) paper and had AI help me make it sound more professional. I submitted it to Reviews and Essays, but it was denied due to the use of AI. It's 100% my own work. I just used AI as a live editor. I'm satisfied with what I have now, but it's being rejected. How do I overcome this?

 

[ElectricBlue]

I wrote a pseudoscientific (I'm not an actual scientist) paper and had AI help me make it sound more professional. I submitted it to Reviews and Essays, but it was denied due to the use of AI. It's 100% my own work. I just used AI as a live editor. I'm satisfied with what I have now, but it's being rejected. How do I overcome this?

You can't, unless you purge every element of the AI edit. The site's policy on AI is that, if you use AI tools in any way (including Grammarly, and you blindly accept suggested changes), then it's not all your own work.

The FAQs are explicit on this, and there are dozens of threads in the Authors Hangout on this issue. You've got to rework it yourself, and take the "AI-ness" out of it.

 

[darwin1859]

I wrote a pseudoscientific (I'm not an actual scientist) paper and had AI help me make it sound more professional. I submitted it to Reviews and Essays, but it was denied due to the use of AI. It's 100% my own work. I just used AI as a live editor. I'm satisfied with what I have now, but it's being rejected. How do I overcome this?

According to the site owner, no one here knows what triggers the AI flagging, but by your own admission, your article violates the AI rules. And the AI rules seem to be interpreted strictly on Literotica.

If you have a version you wrote 100% (before you asked for AI assistance), then I would suggest that you have AI tell you the kinds of edits you could make to make it better and then you write the edits, so you can honestly say that the writing is 100% yours. Be clear in your prompt that you don't want AI to make any wording changes, just flag points where the article could be better and how it could be better. This may, or may not trigger the AI rules, although it should be acceptable.

Or publish your essay elsewhere. Very, very few people read Reviews and Essays, so it doesn't really matter if almost no one reads it here, or no one reads it elsewhere?

Or write something from scratch without using AI.

 

[PennyThompson]

I wrote a pseudoscientific (I'm not an actual scientist) paper and had AI help me make it sound more professional.

I don't mean this in a glib way, but genuinely...

If you couldn't spend the time to write it yourself, why would you expect others to spend their time reading it?

Write your own stories with your own words, not the words of an LLM. If you want to write in an academic style, go for it! Read some journal articles to get a sense of that structure and flow. Take your time to re-read and self-edit your first draft. You can visit the Editors Forum to find a human volunteer editor or beta reader.

It doesn't need to be perfect... it just needs to be human

 

[yano2mch]

I've said before i think the rules are too strict. Getting flagged because you moved a comma, did spellchecking/correction or the like shouldn't disqualify you. And yet it is...

According to the site owner, no one here knows what triggers the AI flagging, but by your own admission, your article violates the AI rules. And the AI rules seem to be interpreted strictly on Literotica.

I can wager a guess.

One method to speed up LLMs is through a draft model, a weaker model that makes guesses of several words at a time, then the main model does a verification pass which is 6x-10x faster and easier. Each token will have a likely/correct next word in a percentage and if that percentage is over a threshold it's treated as correct, which looks like it's 70% or higher on the 'most likely next word'. If not it drops the draft at that point and has the main model fix the word before passing it back to the draft model.

Now, skip out the drafting model and just shove the whole text through for verification/likely next word step, and add those numbers up. If the percentage/average is over a threshold, Poof! was done by AI. Though maybe it only considers the top 10-20 most likely next words, or words adding up to 90% of the most likely possibilities (which can be a lot, or just a few).

---

Solution? Make your work less perfect. Throw out most punctuation, add random upper/lowercase letters in, misspell multiple words in various places. If being correct and doing a good job disqualifies you, then be less perfect. Throw random formatting in for good measure. Do some some double words, don't use smart quotes and use ASCII only. Throw-in random-hyphens. Etc etc.

 

[ElectricBlue]

Solution? Make your work less perfect. Throw out most punctuation, add random upper/lowercase letters in, misspell multiple words in various places. If being correct and doing a good job disqualifies you, then be less perfect. Throw random formatting in for good measure. Do some some double words, don't use smart quotes and use ASCII only. Throw-in random-hyphens. Etc etc.

This is not good advice. There are plenty of writers here who submit extremely high calibre copy, free from errors, as perfect as they can get it.

The trick to not being rejected for suspected use of AI is to write content that doesn't read like it's AI. That doesn't mean submitting imperfect junk, it means not writing like the junk in the first place.

 

[yano2mch]

This is not good advice. There are plenty of writers here who submit extremely high calibre copy, free from errors, as perfect as they can get it.

The trick to not being rejected for suspected use of AI is to write content that doesn't read like it's AI. That doesn't mean submitting imperfect junk, it means not writing like the junk in the first place.

I agree. But it's good work that makes it look like AI. Thus admitting, AI output is probably pretty good, or that the rules picked up for likely following words is more-or-less accurate within expectations of pattern matching.

Many rules put in place that have good intentions will have unwanted side effects. For Tesla cars you 'HAD' to have your hands on the wheel, so they started just tying weights to the steering wheel so it detected hands were there. A camera would look to see if your eyes were open and looking so they started making sleeping bands that looks like eyes are staring forward. You're just putting barriers up that has obscure rules to follow, and throwing in randomness or garbage is the only answer.

"The sky was blue" would probably come back as AI written, while "The sky was mayonnaise" wouldn't because it wasn't in the top 10-100 words. Hell using html tags consistently would flag you as AI. every newline has a <p> block, and the end of the line/paragraph consistently has a </p> block, so the AI would naturally want to write using the <p> and </p> blocks.

I say, don't force a reason to put in lower quality work. You're just going to make people throw up their hands and gimp their own works, or never submit them.

 

[benfan]

I wrote a pseudoscientific (I'm not an actual scientist) paper and had AI help me make it sound more professional.

If someone writes an essay and then uploads it to AI with the prompt: "Make this sound more professional," the output—even if subsequently re-edited—is no longer really their own work.

If they upload it and ask: "Suggest what types of changes would help this sound more like a professional scientific paper," and then composes those changes themselves, I'd call that their own work.

Before submitting my own story I was worried, based on what I'd read here, about it getting flagged as AI. I'm writing in a quirky style—what I'd guess some people might suspect to be AI, even though it's really a human attempt to emulate British prose of ~125 years ago—and I use lots of em-dashes too! The whole kerfluffle about em-dashes being a tipoff for AI alerted me to what handy bits of punctuation they really are.

But it sailed right through. So I have no idea what triggers the reported false-positives.

 

[yano2mch]

But it sailed right through. So I have no idea what triggers the reported false-positives.

I suspect draft model/verification method done via the models; Which means if it's in the range of the 'most likely words to follow' is high enough it will take a sum and just flag it even if it isn't AI. Wrote it a few posts up. Though likely models are going to go diffusion (128-256 tokens at a time similar to image generation) rather than auto-regressive (one word at a time).

I suppose it depends on what model the site is using as a baseline test. Or if it's doing multiple and choosing the worst of multiple models...

 

[Britva415]

How do I overcome this?

Go back to the version you had before you let AI add to/change it. If you think it needs to be added to/changed, then you're already using your own brain and don't need AI and can do it yourself and can submit that. Should work fine.

 

[EmilyMiller]

@theAdventurable sadly the following is very frequent, yet utterly erroneous, advice. Please ignore it

But it's good work that makes it look like AI.

It’s also insulting to authors who write well without AI.

 

[yano2mch]

It’s also insulting to authors who write well without AI.

Not sure why. Each person is good at different things, and one comes to terms with their limitations. If a tool helps one do a better job then maybe they should use the tool.

Or you can just... you know... use a quill pen and paper and try never to make any mistakes at all... And if you want to share a copy, copy it by hand too.

Be a shame of there was technology that made writing easier... checking for errors easier... copying easier...

 

[Britva415]

it's good work that makes it look like AI

Not to me it don't.

 

[yano2mch]

Not to me it don't.

Different models can have vastly different output. As one who RPs, i have a few of my own favorites so far. Long as it doesn't step on the 'i've seen that too many times' those outputs are quite decent, though others... barf or feel wrong overall.

 

[ronde]

Not sure why. Each person is good at different things, and one comes to terms with their limitations. If a tool helps one do a better job then maybe they should use the tool.

Or you can just... you know... use a quill pen and paper and try never to make any mistakes at all... And if you want to share a copy, copy it by hand too.

Be a shame of there was technology that made writing easier... checking for errors easier... copying easier...

There is a huge difference between a "tool" and a "crutch".

A tool is something used by someone with ability to do a job more efficiently. Use of the tool still requires the knowledge of how to do the job. A carpenter will use a powered saw and a nail gun to speed up the process of building a house, but he or she still has to know the right way to build a house. The printing press didn't replace the need for the knowledge of spelling, punctuation, and sentence structure. It just made generation of text faster and made each copy exactly like the one before it, mistakes and all.

A crutch is something used to compensate for a failing of the user. That's what you're trying to call a "tool". You can come to terms with the fact that you don't know how to build a house, but if you try using a powered saw and a nail gun to substitute for that lack of knowledge, the house will most likely fall down. The answer is to learn how to do what you want to do or leave that job to someone who does.

 

[yano2mch]

A crutch is something used to compensate for a failing of the user. That's what you're trying to call a "tool". You can come to terms with the fact that you don't know how to build a house, but if you try using a powered saw and a nail gun to substitute for that lack of knowledge, the house will most likely fall down. The answer is to learn how to do what you want to do or leave that job to someone who does.

Is using a calculator a crutch or is it a tool? I don't understand enough of trigonometry, but i can still use it to calculate the area expected to cover say the backyard for a pool. And it's been said the cosine tables in ships were often wrong and many which may have led to shipwrecks until they were corrected.

I'd LOVE to write my own 3D engine even if it were very simplistic but trigonometry i can't wrap my head around; If an AI or someone else wrote a basic one for me in a way that was usable, I'd probably use it, then write what i COULD myself for the rest of the framework, things I'm good at vs things I'm not. And if something was written that was bad I'd go over it with a fine tooth comb and try to fix it as best i could (or give up and delete the whole thing). Demanding it all be written fully by me when i can't understand the main most important parts though will just result in nothing ever being made.


But how good someone is for writing doesn't seem to be the issue. There's horrid stories by humans and horrid stories by AI, there's good stories by humans, and apparently good stories by AI. You're going to get good results or bad results, or bad results made into somewhat better results from external means (be it a friend giving feedback, paid editor or an AI, or skip it all and just hire a ghostwriter then get the credit while legally being in the right as owner of the work). But you can only bug your friend, mom, or someone else so many times before they just flatly refuse. And even if the AI were to do a lot of the heavy lifting, it's still trial and error so there's some growth there as you are unlikely to write a prompt, then print the result and never look at it. Maybe the personal growth is not as much as say using an oldschool 30's typewriter where you can only type 20 words a minute or it would jam up. Then again my mom used to say, if you wanted to be a good writer, read a lot of books, you'd innately pick up grammar structure plots and many other things that you can't learn elsewhere.

 

[EmilyMiller]

There is a huge difference between a "tool" and a "crutch".

A tool is something used by someone with ability to do a job more efficiently. Use of the tool still requires the knowledge of how to do the job. A carpenter will use a powered saw and a nail gun to speed up the process of building a house, but he or she still has to know the right way to build a house. The printing press didn't replace the need for the knowledge of spelling, punctuation, and sentence structure. It just made generation of text faster and made each copy exactly like the one before it, mistakes and all.

A crutch is something used to compensate for a failing of the user. That's what you're trying to call a "tool". You can come to terms with the fact that you don't know how to build a house, but if you try using a powered saw and a nail gun to substitute for that lack of knowledge, the house will most likely fall down. The answer is to learn how to do what you want to do or leave that job to someone who does.

Totally

There are many bad faith arguments made for genAI - ‘it’s just a tool’ is one of the most egregious. And it’s made by people who want an excuse to use it.

 

[yano2mch]

Totally

There are many bad faith arguments made for genAI - ‘it’s just a tool’ is one of the most egregious. And it’s made by people who want an excuse to use it.

I suppose. Well i'll just say what i say about my life. I don't mind working, but i hate working alone.. Same goes for writing. (Programming, music... eh....)

But if you are using it to translate documents and fix grammar and spelling, isn't it just a tool? The same as those that detect the issues but aren't used by AI? Other than the processing cost, isn't it better to use something that helps?

On the bright side, a lot of the 'free AI' that we've had the last few years is likely going away as it costs too much to run, and a lot of the 'slop' we've seen are people using it just for the sake of using it. When it costs money or effort, then the raised bar may mean it's used more conservatively. Or the slop is so much more obvious and less useful on the lighter models they use to make it work cheaper but expectations of what it can do suddenly plummet.

 

[StillStunned]

But if you are using it to translate documents and fix grammar and spelling, isn't it just a tool? The same as those that detect the issues but aren't used by AI? Other than the processing cost, isn't it better to use something that helps?

I've worked with enough translators to know that translating is a creative endeavour. This is also recognised by copyright law: the basic rule is that the translator owns the copyright to the translation.

So no, using AI to translate isn't a tool. It's theft.

 

[yano2mch]

So no, using AI to translate isn't a tool. It's theft.

Mhmmm... So what if i mapped say spanish words to english words and did a rough translation using the tool, with no AI involved?

Still theft?

Just because it's not technically owned doesn't mean it doesn't have value. Understanding all the nuances of a language certainly makes translating easier. But the ability to enjoy or even use a product you can't read from it's native language shouldn't hamper and limit you.

Besides, theft involves one party losing something. No one loses anything when you are dealing with digital goods.

 

[StillStunned]

Besides, theft involves one party losing something. No one loses anything when you are dealing with digital goods.

Do you understand how copyright works?

 

[yano2mch]

Do you understand how copyright works?

I understand it enough, to know it's utterly broken. Understand i'm talking physical objects vs digital assets.



"To encourage the creation of arts and works, the creator shall for a limited time have the rights to use/sell their works" to paraphrase. But today it's corporations owning everything forever and only wanting to share it via pay-per-view. Not selling you content, but a 'license' as you'll be quite well aware if you've purchased payed for anything on Steam/Epic/GoG, where what you purchased paid for can be yanked at any time.

Originally Copyright length was about 20 years, but it's been extended multiple times, to now Life+70 years, or something akin to 120 years for a corporation. (And if Disney can, they'd extend it another 20 years to keep Micky Mouse out of anyone's hands while also purchasing companies like Pixar amassing even more content)

You have a number of rights when you create something. The usage of Copyright originally was for the printing press, so you hand over the right to copy a said work to another entity, and in theory only they can make a copy of said work. Course the right needs to eventually be returned to the creator owner of said work, though some authors who wanted to put their book in public domain after a couple years were told by the presses they couldn't because even though the book was out of print, they held the permission and wouldn't let it go.

An advantage with books is, the knowledge is out there even with copyright, just who got paid for said works and creation was well established, and copying something physically was a pain anyways.

Not so with current digital media with LICENSES where it's revocable at any time... So I'd say it's the common man who gets screwed at present when they can have their 2000+ TV series or 500 movies ripped from them after they paid for it expecting to have it forever... Or books being removed remotely after being purchased from Amazon...

https://www.foxnews.com/story/orwells-1984-and-animal-farm-deleted-from-kindles
https://www.techspot.com/news/112919-sony-deleting-551-movies-tv-shows-you-bought.html
https://kotaku.com/sony-ps4-ps5-discovery-mythbusters-tv-1851066164

I'd say purchasers are being ripped their rights to own anything. And thus i hold copyright in very low value right now.

 

[yano2mch]

Should i go on?

 

[StillStunned]

Please don't until you've had a creation of yours stolen and sold by someone else.

 

[yano2mch]

Please don't until you've had a creation of yours stolen and sold by someone else.

Apparently i haven't said enough. So i'll continue.

Two people who come up with the same thing at roughly the same time, would both have the copyright, even if made independently (Though how that's handled i don't know, it's likely rare enough to not be an issue, though the RSA encryption with public key encryption was found this way and likely goes more under patent than under copyright)

Now if a work is 25% or more DIFFERENT, then it counts as it's own work. I'm sure this goes very much for more visual designs, so you'd have Star Trek, then you'd have the Kelvin Timeline, which is Star Trek but they had to change it to be different enough.

Now a number of things can't be copyrighted. Individual words, ideas, spoken word in public, etc. Copyright instead covers an INSTANCE of something, not it in general. So Darth Vader scene in A New Hope would be copyrighted to Lucas Films, while Darth Vader as a costume someone made and wears at cons, would not be. (Though he's recognizable enough that they push this to Intellectual property or more likely TradeMark, which is stupid as it just lets companies like Disney do lawfare against you and drain your bank account indefinitely so it's easier to let the franchises die). Though if you had other photographers filming from a different angle, the copyright of the same scene/instance would likely go to the photographer. Same if you had two people take a picture of the statue of liberty side by side, very similar, but not quite the same. Oh the subject is the same, but the actual instance of the photographs are unique.

Thus then we go to AI. AI is trained to find PATTERNS in the input, be it language, image, sound, or motion capture. These patterns then mixed with a bit of static can make NEW patterns, that i would say is at least 25% different from the source. Also AI weights hold a fraction of the training data. Stable Diffusion 1.5 which was released, was about 3Gb in size, while the training data was closer to 10Tb. There's no way to store a fraction of the images.

Now while the AI is not a normal human, it's been ruled the output isn't copyrighted, though they tend to be flaws and fixing said output likely makes it now an original work even if you didn't do it all from scratch.


Now. Are you going to argue that a model trained with say with image of Darth Vader and Harry Potter text, would be stealing from Lucas Films and JK Rowling? Or are you saying an output using any of that content even if it isn't anywhere near close to any instances that they'd made is considered stealing? Or converting text from one language to another is also stealing?