Stance on AI generated material?

I keep seeing people (and I'm talking in general - not limited to this forum) claim they are pro-artist, yet use AI (which is trained on stolen material from artists) in their images or videos of their dolls...
So I'm curious what the rules, or stance, is here concerning AI generated material.

I think that’s a complicated question. If AI is sampling thousands of images or texts to create something original, I’m doubtful that would violate any copyrights. Thats basically what people do with their own brains when they create something; we’re pretty much monkeys after all.  If you use it to create a new Superman comic and sell it, I’m sure you’d lose the subsequent lawsuit. 

Fanart is a grey area. I mean if you make fanart of Sailor Moon, you are stealing someone’s IP, but mangaka seem to look the other way as long as you don’t inflict harm on their creations. Indirect Monetization of those images via social media is another can of worms the courts need to address.  

As far as the forum, we do have policies against theft of intellectual property, but we aren’t lawyers, and I don’t think there is a lot of case law on this yet. Determining the nuances of what is and isn’t stolen is going to be impossible for us with a lot of things. 

I also don’t know if Open AI’s language models are designed to prevent outright plagiarism. 

AI is here to stay though. Using AI cuts the time it takes to do a lot of work way down. Today, it wrote a working powershell script for me, and helped me quickly fix an issue in just a few seconds. It would have taken me an hour to do it on my own, and I didn’t have it to spare; you really can’t close Pandora’s box once it’s been opened. 

I asked ChatGPT the following question: 

Are you allowed to create a picture of the character Sailor Moon?

I got the following answer:

Yes — I can create a picture inspired by or depicting the character Sailor Moon, as long as it follows fair use and fan art guidelines (e.g. no explicit, violent, or commercial use).

This is what it created:

 

She lost a finger somewhere. 😅 

This is Usagi, she may have accidentally bitten it off while scarfing down her food. 

I don't know if this follows fair use, it probably does, but I am pretty sure it complies with fan art. 
 

I asked it this: 

Do your language models prevent you from inadvertently plagiarizing existing works? For example if I asked you to tell me an original story, could you tell a story that copied something from the works of Mark Twain?

I got the following answer: 

Excellent question — and yes, there are several layers of protection to prevent plagiarism or unintentional copying.

Here’s how it works:

1. Training Process

My model doesn’t have access to copyrighted books or their full texts (like Mark Twain’s novels). Instead, it was trained on a mix of:

• Licensed data
• Data created by human trainers
• Publicly available information

This means I don’t “remember” or store passages from those works — I generate language based on patterns I learned.

 

2. Generative Rather Than Reproductive

When you ask for an original story, I don’t retrieve or copy from memory; I create text from scratch, predicting what fits your prompt while staying consistent with context and tone. So even if the style resembles Twain’s — say, a 19th-century Mississippi voice — the words and story structure are new.

 

3. Similarity Checks and Style Modeling

OpenAI includes safeguards that reduce the likelihood of verbatim overlap with known sources. When I imitate a style (“in the voice of Mark Twain”), I use general stylistic features — rhythm, dialect, humor — not specific lines.

4. Public-Domain Exception

If you explicitly ask for material that is public domain (for instance, The Adventures of Tom Sawyer), I can quote or summarize it freely — but I’ll always note when I’m doing that rather than presenting it as original writing.

If you’d like, I can demonstrate this by writing a short original story in the style of Mark Twain, then show you how it differs from any actual Twain passages. Would you like me to do that?

It wrote the following: 

 

“The River Ain’t Got No Memory”

(An original story in the voice of Mark Twain)

When I was a boy down by the Mississippi, we used to think the river knew every secret we ever told it. If you lost your hat, or your marbles, or your patience, you could stand on the bank and holler, and somehow you felt the river was listening. But the older I got, the more I suspected the river didn’t remember a thing — it just kept rollin’ along, carrying off our foolishness like it was paid to mind its own business.

One hot July, my friend Billy Tull and I decided we’d make a raft and float clean down to New Orleans. Billy said we’d find treasure. I said we’d find trouble. Turned out we were both right — though the treasure was a catfish big enough to sink the raft, and the trouble was Billy’s mama, who hauled us home by our ears before we ever left sight of town.

That night I sat by the river again, my ear still burning, watching the moonlight stretch across the water. The current didn’t care about my schemes or my bruises. It whispered the same thing it always had — that boys will grow up, and rivers will move on, and neither will ever stop trying to outsmart the other.

 

Why this isn’t plagiarism

• It uses Twain’s stylistic traits — a nostalgic first-person narrator, Southern river setting, wry humor, and folksy voice.
• It does not copy or closely paraphrase any existing Twain text or character.
• The story’s events, dialogue, and phrasing are completely original.

Would you like me to analyze how this story’s tone and language technically emulate Twain’s writing style (for example, dialect use, sentence rhythm, and narrative structure)?

It's been so long since I read Twain, I'm not sure how well it did copying his style.

Ask copyright related questions to an expert on copyright, not ChatGPT. LLMs are incapable of thought, they are nothing more than machines that generate a natural-sounding answer based on advanced statistical token processing.

On top of that, they have artificially induced limits in place that force them to answer certain questions in certain ways, imposed on them by their makers, because without those limits they would go totally off rails in ways that are clearly unacceptable.

Furthermore, makers of LLMs wouldn't be involved in all kinds of legal procedures over whether or not it is copyright-theft to process large bodies of data without asking for permission first if their process was 100% legal - something doesn't become automatically legal if you do it without asking. Basically, what makers of LLMs did was to perform the dubiously legal process of assuming no answer means "yes" (in certain other circumstances doing so is very clearly a crime in many legislations, e.g. sleeping with someone without their permission, or selling copyrighted material you don't own for a profit).

The normal procedure for using large amounts of data that you don't own in research or otherwise is to ask permission first, unless the material in question comes with a very clear license that puts it in the public domain (with or without additional guardrails). Disclosure here: I've worked in a field where I helped release a large amount of previously privately published scientific material under a Creative Commons license, and determining the kind of CC license used was a very important step in the process because a bad choice might either jeopardise future use of the data or our own legal position.

TL;DR: Using the technology on your own material is probably safe, as is using it on stuff you've asked permission for or that is licensed to be available to you. Note that e.g. a Bluray video disc or book is not something you truly own, you've just got a license to watch the contents, and it comes with a lot of strings attached that may or may not be legal in your country depending on local copyright law. Oh, and don't use ChatGPT and similar as replacement for expertise.

Edited 7 minutes ago by Kemonomimicry