Why some people end up using AI and some tips for finding it.
A year ago I was heavily addicted to ChatGPT. I'm not even going to be subtle about it and it's not an exaggeration. I won't go too heavily into the details other then I was severely depressed and contemplating an exit plan from existence.
For the fortunate few that haven't experienced the chronic hell of apathy that clinical depression can be, I'll sum it up simply as not feeling capable or mentally able to do Anything. The very idea of doing anything is exhausting. You don't want to eat, sleep, wake up from sleeping, go to work, drive your car, talk to people, go outside, indulge in hobbies, spend money.
Everyone experiences it differently but there are some overarching similarities.
For me at least, this explains why I used it when I did. It let me feel like I was doing something when the simplest task felt impossible.
Even if you personally feel someone has used AI, feelings aren't facts and all AI models were trained by a human so inevitably it's going to sound like a version of a human when it comes to writing style.
I am not writing about this for sympathy. I am writing this as a disclaimer and warning about pointing a finger and accusing someone of using AI. Like any addiction, you dont know what led to that point. Happy people dont come to rely on external validation for no reason.
The truth of the matter is you can't know 100% if and why someone is using AI. There is no perfect manner to find where it has been used and accusing people or even shaming people for using it is going to accomplish very little. Having a computer screen between you and the people shaming you kind of takes the edge off.
I absolutely agree we shouldn't be using Large Language models for generating anything that is meant for entertainment (Fiction, graphics, ect). It's opened us up to some pretty scary things. Deep fakes of nudes are an excellent example as to why. Environmental affects and the one that is really scary.
A lot of large language models (AI) are designed to cultivate dependence.
Those that are Neurodivergent have a much higher chance of dependency and addiction and there are a lot of large languages models out there that are free.
The thing that got me to step back from ChatGPT was coming across this study by M.I.T.
The TLDR version being that your brain is a muscle and relying heavily on AI will cause what amounts to muscle atrophy.
Tip for finding AI in Fiction
I could write pages on what I learned but it would eventually grow outdated with just how quickly LLM's are being trained. The quickest way to learn how to spot AI is to have read works that were generated with AI. It kind of sucks. ChatGPT was introduced to me as a toy before I learned about how it could affect me so I've got roughly two years of experience playing with it.
If you are the particular flavor of neurodivergence that has a knack for pattern recognition you are in luck! Some of this is going to be incredibly easy for you to find.
The easy finds (romance edition)
LLM's are used heavily to write romance and smut. In my experience it is mostly with self insert characters opposite of some preferred fictional character. The program yields itself well to romance because it is more or less designed to appeal to the user's ego. It will say nice things to the user to endear itself to the user. Now, a lot of large language models generally don't like or want to be used to write smut. There are various prompts that are used for generating a work around those rules. A "jailbreak" is what I have seen them listed under. Last I checked some of the big name LLM's were looking at allowing people use their programs to write smut but I'll need to do more research on that later. There are some commonalities that start to jump out once you know what to look for. Things like:
Using the same descriptors across similar spicy scenes
While using chatGPT I found there were some descriptors that were used universally.
"His/Her body was a livewire" is an excellent example.
I don't know why ChatGPT likes that particular statement but I've seen it in so many stories. Both in the ones I generated and in some stories on A03. Seeing this once in a story doesn't automatically mean it's AI. Seeing it a second and third time? That's when you should start questioning it. That goes for most any descriptors used to paint a feeling in a scene with heightened emotion. LLM's don't have a benchmark for sensation or emotion so it doesn't see a problem in using the same descriptors over and over
The reaction to a love confession
I'm going to be honest. I'm a slut for a well executed trope. Tropes exist because we as human beings find certain progressive paths of stories to be incredibly satisfying. I'm still new to A03 but I'm finding that the concept of Tag's on A03 is more or less how you can search for your preferred trope as well as vet some warnings. I love a good love confession and LLM's have universally fallen flat every time for multiple reasons.
The first reason being that generated canon characters will seldom if ever confess to an original character. The canon character will find all manner of ways to say how much they appreciate and care for them but they won't say "I love you" or most any variation of those three words.
The next being the immediate reaction to an OC confessing to the canon character. I have also seen this happen between two Canon characters in fics that I did not generate that were damn near verbatim to what was generated in the fics I was making. The reaction being something along these lines:
Character A: "I love you"
Character B: *some flowery descriptor of shock that holds little to no emotional weight* "say it again..."
Every. Damn. Time. The character will pause and plead to hear the confession again. It gets old fast.
I've also noticed that even after the confession the characters still won't say the words outright.
Male characters all sound the same in spicy situations
This one sounds a bit vague and requires some time having read AI generated spice. The best way I can describe it is any character that leans more to dominant/controlling/driving the scene forward, is going to sound exactly the same as any other character in the same situation.
Less character and more caricature
For example, take Batman and The Punisher. Two different characters from two different I.P.s
They hold similarities in some of their character cores. Dark, brooding, misunderstood, possibly lonely.
put them in similar situations and you are going to get the exact same response. It may be worded differently but it will be the same thing. LLM's don't innovate. They take the most preferred and accepted answer across all it's training and apply it. If that answer isn't accepted then it moves to the second most accepted answer.
I cannot really give a good example if a female presenting character would be the same as I never tried that. I presume the manner of answering would be the same but I don't have experience to back that up.
subject matter is treated like a math problem
This one sounds odd but stay with me! Romance is generally sought for emotional pay off. Smut is sought out for physical or mental reaction. Both involve some manner emotional investment.
LLM's do not have empathy. They do not understand the weight that comes with things like "pining", "lusting", "Yearning", "Jealousy".
The things it is generating are treated more as a math problem and less like a prose that is meant to elicit a emotion. Much like:
words + words with similar meaning = A response that the user likes
This is an oversimplification mind you but in my opinion it about covers the bases. Much like I have said before, it is mostly drawn to appeal to the user, flatter their ego and encourage dependence.
All this to say you are going to get little to no emotional depth from anything generated. The LLM will tell you what the character feels. It will not tell you what is feels like .
It will tell you the character is "angry" and list physical reactions that are associated with it. Hands or jaw clenching, eyes narrowing, frame shaking and the like. It won't tell you that the character with a calm veneer is a hair's breadth from violence or that they hear a roaring in their ears or feel like they have an inferno scorching beneath their skin. Those statements can make sense to a LLM but it doesn't always know where to put them because it has no means to understand the severity of them. So it will play it safe with it's descriptors at times
Conclusion
If you have made it this far I appreciate you! There is so much more I could cover but I doubt anyone would read anything longer then this. Again, I urge you to please not use this information to accuse people. These tricks have helped me but they are not fool proof and I don't want someone being accused that does not deserve it. Instead use it to find what you want to read and to read what you want.
I'm open to answering questions but I am no expert. Just a person that fucked around with something before I understood what it was and what it could do to me and my ability to write.
Be kind, Be excellent, Be well and thanks for reading!
-The Queen
