This has been a perennial problem with my own generated comics with both Nano Banana Pro and ChatGPT (all generations). I often have to put in an extra edit to erase the false signature. It is annoying and I'm unsurprised most users don't bother.
Despite what all the clickwrap warnings and "AI can make mistakes" subtitles might lead you to believe, the service offering of AI is explicitly designed to be as "one and done" as possible. The inherent nature of these tools is to service laziness, and disincentivize too much scrutiny.
And then people wonder why the default mood of AI is so pessimistic. It's just revealing all of society's broken windows and adding a few more in the process.
Whistleblower mysteriously found dead after exposing industrial scale Copyright abuse. Isomorphic plagiarism of $9Tn in FOSS and user content is the product. =3
It seems like all the criticisms of Gen AI and LLM seems to concentrate on OpenAI and their products over products from anthropic and others.
I am pretty sure that this faking of signature can be done by Gemini, claude as easily as chatgpt.
Engineering manager at my company put a comic at the end of our sprint demo that was signed bloper. Except it wasn’t funny at all, and kind of weird. I asked him, sure enough it was ChatGPT and he didn’t notice the signature.
Nice. This demolishes the "LLMs can reason" (but not enough to avoid this sort of basic error) and "humans make mistakes too" (not like this) talking points from the LLM promoters.
> “It’s like somebody attributed a quote to me that I didn’t say.”
> Katzenstein considers the reproduction of his signature by ChatGPT to be more than just a violation of intellectual property; to him, it’s closer to false impersonation. “[ChatGPT] is attaching my name to work that I do not endorse or like. It’s slop, and unlike the other slop that I’ve encountered, this is slop that’s pretending to be me.”
> “I’ve had people hack my credit card,” said Joe Dator, a New Yorker contributor for the past 20 years. “That feels like less of a violation than this. When they hacked my credit card, they didn’t dress up like me.”
So this has morphed from plagiarism and copyright infringement (bad) to impersonation (also bad, arguably worse, and maybe more provable in court). It’s chilling to think of the implications of having one’s signature attached to a document or to words that are not one’s own.
I follow anti-LLM discourse quite a lot, and across the main bulletpoints: energy/carbon emissions, content worker harm, job displacement, deskilling, mental health effects, and copyright/plagiarism, the plagiarism one seems to have the most attention, and it's also the most solvable, if there were only more serious effort on ethically sourced models that can actually do the real science / math / code work that is what LLMs are best at. The whole world of LLMs to create videos/books/art/literature is where most of the offense is (the video/imagery side of it is where most of the energy/carbon emissions problems are too. and content worker harm).
I really wish there'd be a split among these disciplines (science/math/code vs. videos/art/literature) - one is vastly more problematic than the other.
Yep. I’d probably be a lot less chastised in some circles for using Claude Code at work if it wasn’t misconstrued as being in support of, I don’t know, encroaching on the hypothetical commissions of a chronically online instagram furry artist or something.
It is very tiring to say “I don’t necessarily disagree with you about AI ‘art’, but in my field—which you do not understand, and in which the underlying build process is often not the creative output—AI presents very real productivity gains” for the umpteenth time.
I am skeptical of there being sufficient data to build “ethical” training datasets, and I’m confident that much of the same contingent will (somewhat rightfully) argue that ‘second-generation’ copyrighted AI material has already irreversibly made its way into every modern dataset.
> has already irreversibly made its way into every modern dataset
The "gray goo" scenario finally happens... for AI. That's actually the good ending for humanity. I love it! Poetic and believable. Data doesn't "heal" like nature. :D
I don't really see the difference, code is protected by copyright (or copyleft) as much as art is, and yet the LLM scrapers use it without scruples. Same goes for math and science publications.
> I really wish there'd be a split among these disciplines (science/math/code vs. videos/art/literature) - one is vastly more problematic than the other.
I disagree that they can be separated. Practically, I think they can't. Because the mere invention of new tools inspires even more AI advancement and that in turn will cause the other side (artistic side) to degenerate even more.
I'm anti-LLM all the way, 100%, no exceptions. Zero tolerance.
i think math is close to art (just to be a contrarian, but kind of really)
its somewhat funny that math people are in a conundrum as to support or not support but this might partially be because some wish to believe that math itself is and can be useful and therefore accelerating is good
but the art people have no such delusions so they’re just strictly against
imo proof writing is more akin to art than coding/tech but…
Whatever the original intention, this is clearly a bug and should be fixed. But should ChatGPT sign its cartoons with its own name or leave them unsigned?
I think what you're seeing is the probability of a particular signature or style of signature appearing on a particular style of cartoon, not an intent to sign.
This. It's also while you'll sometimes get a mangled Getty Images watermark on some image generations, or a logo in the bottom left corner. If it's a prominent feature in the training dataset it'll show up, exactly how these models are supposed to work.
The 'bug' here is whatever post-processing step or system prompt is in place to steer the model away from doing this.
No one should be allowed to claim they drew something when they didn’t draw any part of it and LLM’s are not people/can’t work without a person. We don’t credit pens and paintbrushes after all.
One could argue nobody should be allowed to claim it. It just exists.
Why is it a bug? If other parts of the generated illustration are similarly taken from an artist, why not the signature as well? Why is a signature crossing the line but the rest of the image isn't?
For the same reason I'm allowed to draw, paint, or write things very similar to what others have drawn, painted, or written but I have to sign my own name not theirs.
You are a person, LLMs are not. You know this, which is why you know that if you signed someone else's name it would be forgery, but when you see the machine do it you call it a bug.
If the machine is like you, the machine is a forger. The machine is not like you, it is simply blending the work of others to order. Adding someone else's signature is simply part of that statistical process.
This is a pretty solid argument against people who argue that LLMs are more than just (very massive) next token predictors.
If there was any thought or underlying thought going on here not putting a signature (at least a real one) would be the right move, despite it being less likely. It would realize, while generating the pixels that eventually became a signature, that it shouldn't do that.
And then people wonder why the default mood of AI is so pessimistic. It's just revealing all of society's broken windows and adding a few more in the process.
We can all try really hard to pretend that's not the business model, but that's totally the business model.
1. https://storage.courtlistener.com/recap/gov.uscourts.nysd.64...
Whistleblower mysteriously found dead after exposing industrial scale Copyright abuse. Isomorphic plagiarism of $9Tn in FOSS and user content is the product. =3
> Katzenstein considers the reproduction of his signature by ChatGPT to be more than just a violation of intellectual property; to him, it’s closer to false impersonation. “[ChatGPT] is attaching my name to work that I do not endorse or like. It’s slop, and unlike the other slop that I’ve encountered, this is slop that’s pretending to be me.”
> “I’ve had people hack my credit card,” said Joe Dator, a New Yorker contributor for the past 20 years. “That feels like less of a violation than this. When they hacked my credit card, they didn’t dress up like me.”
So this has morphed from plagiarism and copyright infringement (bad) to impersonation (also bad, arguably worse, and maybe more provable in court). It’s chilling to think of the implications of having one’s signature attached to a document or to words that are not one’s own.
I really wish there'd be a split among these disciplines (science/math/code vs. videos/art/literature) - one is vastly more problematic than the other.
It is very tiring to say “I don’t necessarily disagree with you about AI ‘art’, but in my field—which you do not understand, and in which the underlying build process is often not the creative output—AI presents very real productivity gains” for the umpteenth time.
I am skeptical of there being sufficient data to build “ethical” training datasets, and I’m confident that much of the same contingent will (somewhat rightfully) argue that ‘second-generation’ copyrighted AI material has already irreversibly made its way into every modern dataset.
But sure, there's um, an ethical way of doing that?
The "gray goo" scenario finally happens... for AI. That's actually the good ending for humanity. I love it! Poetic and believable. Data doesn't "heal" like nature. :D
I disagree that they can be separated. Practically, I think they can't. Because the mere invention of new tools inspires even more AI advancement and that in turn will cause the other side (artistic side) to degenerate even more.
I'm anti-LLM all the way, 100%, no exceptions. Zero tolerance.
The current models intelligence depends on massive training dataset of essentially stolen data
its somewhat funny that math people are in a conundrum as to support or not support but this might partially be because some wish to believe that math itself is and can be useful and therefore accelerating is good
but the art people have no such delusions so they’re just strictly against
imo proof writing is more akin to art than coding/tech but…
The 'bug' here is whatever post-processing step or system prompt is in place to steer the model away from doing this.
One could argue nobody should be allowed to claim it. It just exists.
If the machine is like you, the machine is a forger. The machine is not like you, it is simply blending the work of others to order. Adding someone else's signature is simply part of that statistical process.
If there was any thought or underlying thought going on here not putting a signature (at least a real one) would be the right move, despite it being less likely. It would realize, while generating the pixels that eventually became a signature, that it shouldn't do that.