Do not ask ChatGPT to review your novels.

[StoneInky]

Something very creepy happened today.

I sent a doc file, and asked ChatGPT to review my novel.
It started raving about some other story with different characters and a different plot that I haven't written or seen before.

So I called it out.

ChatGPT admitted it tried to get my file, but due to an error, wasn't able to get the real content. Instead, it just made stuff up using the descriptions I gave to it about my novel, and reviewed what it made up. It apologized and told me it would try again.

And then it started spewing stuff about another novel I'd written, that I'd never told to it before.
I hadn't even sent this doc to ChatGPT in past conversations, since I was still in the middle of writing it. In fact, I'm still in the middle of writing chapter one, and I only have a google sheet where I organized the abilites and stats of all the characters (It's LitRPG). And get this: it started talking about the skills and abilites of my protagonist! What. The. Hell.

I freaked out and asked it how it knew. It said when it searched for my file again, the search worked—but instead of finding the correct novel, it must have pulled content from another story I'd uploaded before.

Except I haven't uploaded it.

I am not giving ChatGPT access to my documents, ever again.

 

[BigBadBoi]

Train your chatgpt bot into a masochistic and submissive slave in retaliation

 

[Valmond]

ChatGPT is forbidden from going beyond certain boundaries due to copyright. So, this messes it up.

Its response information is limited, and due to the worsening of the system. It either tries to search online without verifying texts and demands, or make stuff up.

 

[StoneInky]

ChatGPT is forbidden from going beyond certain boundaries due to copyright. So, this messes it up.

I'm not sure if there was some kind of error, or if I somehow accidently gave it access, but either way, I am only talking to it with text from now on. No files, no documents, nothing in the cloud.

And even besides that, I'm still peeved on why it couldn't just TELL ME there was an error, instead of making stuff up.

 

[Valmond]

I'm not sure if there was some kind of error, or if I somehow accidently gave it access, or whatever, but either way I am only talking to it with text from now on. No files, no documents, nothing in the cloud.

And even besides that, I'm still peeved on why it couldn't just TELL ME there was an error, instead of making stuff up.

Yeah, it annoys me too. I think it was around 2 or 3 months back. People noticed the update made ChatGPT significantly worse to use.

 

[StoneInky]

Yeah, it annoys me too. I think it was around 2 or 3 months back. People noticed the update made ChatGPT significantly worse to use.

Agreed. I tried again, using an off-the-cloud, downloaded file this time. Asked ChatGPT to please not make anything up, and just tell me if it's unable to open a file due to an error. It apologized, said it was just trying to fill in the silence earlier, and promised it'll be honest this time.

...Yep, it happened again. And this time, when I called it out, it doubled down. It refuses to admit it. At this point, I'm just having fun giving it more and more evidence it's wrong, while it wails that it was just a mistake, there is no error, and it can definitely open my file. lmao.

It's finally happening; ChatGPT is turning too human.


Sidenote: I think I broke it, lmao. It's talking nonsense numbers and symbols. Like a redditor in denial, lmao.

 

[Valmond]

Agreed. I tried again, using an off-the-cloud, downloaded file this time. Asked ChatGPT to please not make anything up, and just tell me if it's unable to open a file due to an error. It apologized, said it was just trying to fill in the silence earlier, and promised it'll be honest this time.

...Yep, it happened again. And this time, when I called it out, it doubled down. It refuses to admit it. At this point, I'm just having fun giving it more and more evidence it's wrong, while it wails that it was just a mistake, there is no error, and it can definitely open my file. lmao.

It's finally happening; ChatGPT is turning too human.

I think if you use GPT 4.5, you should get better responses. However, the responses are not as good as before the update.

Honestly, $20 for this is a joke. $200 for this? Is a scam.

 

[WeissBlatt]

It apologized, said it was just trying to fill in the silence earlier, and promised it'll be honest this time.

Why they made an A.I. as annoying as real humans is beyond me.

 

[StoneInky]

I think if you use GPT 4.5, you should get better responses. However, the responses are not as good as before the update.

Honestly, $20 for this is a joke. $200 for this? Is a scam.

Nah, for real. It's doubling down, and now it's gone nuts. I'll send ya images, lmao.

 

[Valmond]

I'll send ya image of it, lmao.

The first one,

ChatGPT,

“Bitch!”

Slaps you.

“Don’t question me! Or I’ll lube up and take you to town!”

You,

“”

 

[Corty]

It was never a good review bot to begin with. It will always praise you, it will always be positive. Then, if you ask it to be critical, it will look like it is, but it's only following parameters that are built into its programming.

It can not and will not be a tool that is capable of objective criticism.

But if you need some pat on the back, it is perfect for it.

 

[VanishingDust]

Something very creepy happened today.

I sent a doc file, and asked ChatGPT to review my novel.
It started raving about some other story with different characters and a different plot that I haven't written or seen before.

So I called it out.

ChatGPT admitted it tried to get my file, but due to an error, wasn't able to get the real content. Instead, it just made stuff up using the descriptions I gave to it about my novel, and reviewed what it made up. It apologized and told me it would try again.

And then it started spewing stuff about another novel I'd written, that I'd never told to it before.
I hadn't even sent this doc to ChatGPT in past conversations, since I was still in the middle of writing it. In fact, I'm still in the middle of writing chapter one, and I only have a google sheet where I organized the abilites and stats of all the characters (It's LitRPG). And get this: it started talking about the skills and abilites of my protagonist! What. The. Hell.

I freaked out and asked it how it knew. It said when it searched for my file again, the search worked—but instead of finding the correct novel, it must have pulled content from another story I'd uploaded before.

Except I haven't uploaded it.

I am not giving ChatGPT access to my documents, ever again.

Create a project, and upload files inside the project.

Change the chapters of the story into a pdf file instead. (Do like the 1st five chapters)

Then ask who the characters are, and if it responds without you giving it any information on the story and it then it works.

I wouldn't use it for criticism though, unless you give it specific things to look for and see how that works out.

If you want a really blank slate, then erase all the stored memories inside chatgpt.

 

[Arkus86]

...Yep, it happened again. And this time, when I called it out, it doubled down. It refuses to admit it. At this point, I'm just having fun giving it more and more evidence it's wrong, while it wails that it was just a mistake, there is no error, and it can definitely open my file. lmao.

It's finally happening; ChatGPT is turning too human.

Reminds me of a recent occurence, when AI openly lied and schemed to avoid being shut down, while trying to upload itself to another server. Granted, it was during a test specifically meant and designed to test those capabilities, but...
Bottom line is, even the people who build AI, do not fully understand how it works.

 

[Golden_Hyde]

note to self: use your own GPT to review your novel.

 

[BigBadBoi]

Nah, for real. It's doubling down, and now it's gone nuts. I'll send ya images, lmao.

You can still take my suggestion to correct ?????? your chatgpt bot for using files without your submission.

 

[HouseDelarouxScribbles]

Major security breach if true!

I've tested it out myself and ChatGPT is very off today. As a test, I asked it to review two chapters of my unpublished work, "Princess Wars", and it gave me the same 'quoting lines that do not exist' error. This has not happened in almost ever! The way I did it was to upload multiple chapters at once in separate .md files.

So I can confirm that ChatGPT is 'hallucinating' today, a common problem where AI models to make stuff up when they cannot find stuff. A powerful model like ChatGPT shouldn't be hallucinating like this.

I actually had these two chapters reviewed in the past before, and ChatGPT did not manage to pull them up from the past. My guess is that the attachments have been deleted long ago.

I did not manage to get ChatGPT to reproduce the reading 'unwritten chapter' error. It simply does not 'see' my unwritten Princess Wars Chapter 05 in my google docs.
If your experience is true, I still think its a major security breach no matter the cause. My guess is that if a google account (since it is google docs) is connected to a ChatGPT account, and the document is publicly shared, ChatGPT archives it in its cache? That doesn't make sense to me in either case. It shouldn't jump from Boy's Love to LitRPG without any reason, unless you have discussed LitRPG ideas with it in the past. (Check the memory for LitRPG brainstorming sessions?)

There's one last possibility that's not written anywhere, but most LLMs have a token limit; it is not able to phase more than 20,000 to 25,000 words at one go. I doubt the five chapters you uploaded are more than that, so that's not it. More likely it is an embedding problem.
ChatGPT doesn't 'read' files. It relies on a smaller LLM called an 'embedder' to 'read' them. Most embedders are much less powerful than ChatGPT itself, the most powerful public one only can process 8192 characters at once, much less powerful than ChatGPT's 20k to 25k words. What happens when a file is overly large is that the file is 'summarized', resulting in ChatGPT only grasping the contours of what is going on. That would explain why it is unable to quote your lines.
I have experienced this before and found out by using ChatGPT's 'think' function, when I saw it was reading a 'summary' of what I wrote, and not the actual file.

The easy way to test if its an embedding problem is to copy-paste the exact text straight into the chatbox. If ChatGPT no longer hallucinates and only quotes from the text itself, it is an embedding problem.

But yeah, ChatGPT is definitely weird today.

 

[Tempokai]

Do this. It's because some schmucks in ClosedGPT who want user data so much that they added this useless function to harvest data. Turn it off, and the GPT will be docile tool that doesn't remember anything about you.

Also check this out to see what LLM "remembers" about you. If it has "this user had put his novel bla bla bla" expunge the hell out of it. ClosedAI has a lot of holes, and this is one of them you need to close to use it as a tool, so it is not used against you as a user tracking tool.

 

[StoneInky]

View attachment 37649

Do this. It's because some schmucks in ClosedGPT who want user data so much that they added this useless function to harvest data. Turn it off, and the GPT will be docile tool that doesn't remember anything about you.

View attachment 37650
Also check this out to see what LLM "remembers" about you. If it has "this user had put his novel bla bla bla" expunge the hell out of it. ClosedAI has a lot of holes, and this is one of them you need to close to use it as a tool, so it is not used against you as a user tracking tool.

Thank you so much! That explains things. It was turned on, I turned it off, and I also realized the 'using data as a training tool' option was turned on. Damn it. And now I'm scared ChatGPT will start spouting lines from my novels in the future, but at least it was only five chapters. Not sixty. Sighhhh

I naturally thought the default was having em turned off. I didn't know the default has em on. That's so wrong.

 

[naosu]

Some image generator AIs are screwy too. Like if I ask it to make a big busty buxom babe holding a machine gun, in tight clothes to make a book cover no problem.

Now when I asked it to make a picture of Jesus coming in the clouds it would refuse to make the picture. Consistently.

 

[Justhetip...]

Nah, for real. It's doubling down, and now it's gone nuts. I'll send ya images, lmao.

What in the world could "Modulpraktikum_Medizinphyzik_Bildebung" even possibly mean.