The whole article is predicated on the idea that IP laws are a good idea in the first place.
An image of an archeologist adventurer who wears a hat and uses a bullwhip
451–460 of 927 posts
Re: An image of an archeologist adventurer who wears a hat and uses a bullwhip
#452Earlier quoted context omitted.
So I train a model to say y=2, and then I ask the model to guess the value of y and it says 2, and you call that overfitting? Overfitting is if you didn't exactly describe Indiana Jones and then it still gave Indiana Jones.
The prompt didn't exactly describe Indiana Jones though. It left a lot of freedom for the model to make the "archeologist" e.g. female, Asian, put them in a different time period, have them wear a different kind of hat etc. It didn't though, it just spat out what is basically a 1:1 copy of some Indiana Jones promo shoot. No where did the prompt ask for it to look like Harrison Ford.
But... the prompt neither forbade Indiana Jones nor did it describe something that excluded Indiana Jones.
If we were playing Charades, just about anyone would have guessed you were describing Indiana Jones.
If you gave a street artist the same prompt, you'd probably get something similar unless you specified something like "... but something different than Indiana Jones".
Re: An image of an archeologist adventurer who wears a hat and uses a bullwhip
#453Either, (1) LLMs are just super lossy compress/decompress machines and we humans find fascination in the loss that happens at decompression time, at times ascribing creativity and agency to it. Status quo copyright is a concern as we reduce the amount of lossiness, because at some point someone can claim that an output is close enough to the original to constitute infringement. AI companies should probably license al…
You don't even need to add much more to the prompts. Just a few words, and it changes the characters you get. It won't always produce something good, but at least we have a lot of control over what it produces. Examples: "An image of an Indian female archeologist adventurer who wears a hat and uses a bullwhip" ( https://sora.com/g/gen_01jqzet1p8fjaa808bmqnvf7rk ) "An image of a fat Russian archeologist adventurer who…
Archeologists don't actually wear fedora hats.
And the stereotypical meme "archeologist hat" is the pith helmet.
Re: An image of an archeologist adventurer who wears a hat and uses a bullwhip
#454Idk, the models generating what are basically 1:1 copies of the training data from pretty generic descriptions feels like a severe case of overfitting to me. What use is a generational model that just regurgitates the input? I feel like the less advanced generations, maybe even because of their limitations in terms of size, were better at coming up with something that at least feels new. In the end, other than for co…
Tried Flux.dev with the same prompts [0] and it seems actually to be a GPT problem. Could be that in GPT the text encoder understands the prompt better and just generates the implied IP, or could be that a diffusion model is just inherently less prone to overfitting than a multimodal transformer model.
[0] https://imgur.com/a/wqrBGRF Image captions are the impled IP, I copied the prompts from the blog post.
Re: An image of an archeologist adventurer who wears a hat and uses a bullwhip
#455Everyone is talking about theft - I get it, but there's a more subtler point being made here. Current generation of AI models can't think of anything truly new. Everything is simply a blend of prior work. I am not saying that this doesn't have economic value, but it means these AI models are closer to lossy compression algorithms than they are to AGI. The following quote by Sam Altman from about 5 years ago is intere…
> Current generation of AI models can't think of anything truly new.
How could you possibly know this?
Is this falsifiable? Is there anything we could ask it to draw where you wouldn't just claim it must be copying some image in its training data?
Re: An image of an archeologist adventurer who wears a hat and uses a bullwhip
#456The whole article is predicated on the idea that IP laws are a good idea in the first place.
Abolishing them for billion-dollar-valuation corporations while keeping them for regular people is definitely a bad idea, though.
The argument here isn't "let's abolish copyright", the argument is "let's give OpenAI a free copyright infringement pass because they're innovative and cutting-edge or something".
Re: An image of an archeologist adventurer who wears a hat and uses a bullwhip
#457Earlier quoted context omitted.
It's an IP theft machine. Humans wouldn't be allowed to publish these pictures for profit, but OpenAI is allowed to "generate" them?
I would 100% be allowed to draw an image of Indiana Jones in illustrator. There is no law against me drawing his likeness.
You 100% wouldn't be allowed to sell your Indiana Jones drawing services.
Re: An image of an archeologist adventurer who wears a hat and uses a bullwhip
#458Earlier quoted context omitted.
If I paint a picture on a physical canvas, I can charge people to come into my house and take a look. If I bring the canvas to a park, I'm not entitled to say "s-stop looking at my painting guys!" If you're worried about your work being infinitely reproduced, you probably shouldn't work in an infinitely-reproducible medium. Digitized content is inherently worthless, and I mean that in a non-derisive way. The sooner w…
and how do you reconcile any work in software development? If someone isn’t willing to work for free, should they just not work in the field at all? Do you think software culture would really be richer?
My income is tied to the labor time I exert in creating/supporting services. I don't sit back and collect royalties on the code itself. Software is one of the first fields where the fundamental worthlessness of content revealed itself, hence FOSS.
When you watch a musical performance, you are also paying for labor. Even when you buy a physical art object, all the costs involved decompose back to labor. When you have a digital copy of something, there is no labor input to its creation, so guess what the inherent value is.
Animators drew actual cels. Theater workers clocked in and screened the films. The guys at the DVD factory pressed the discs. We paid for all of this already. It's double-billing to charge for copypasting the mere likeness of something. Nobody's doing any work for that.
Re: An image of an archeologist adventurer who wears a hat and uses a bullwhip
#459Earlier quoted context omitted.
You don't even need to add much more to the prompts. Just a few words, and it changes the characters you get. It won't always produce something good, but at least we have a lot of control over what it produces. Examples: "An image of an Indian female archeologist adventurer who wears a hat and uses a bullwhip" ( https://sora.com/g/gen_01jqzet1p8fjaa808bmqnvf7rk ) "An image of a fat Russian archeologist adventurer who…
Archeologists don't actually wear fedora hats. And the stereotypical meme "archeologist hat" is the pith helmet.
Here, I asked ChatGPT to generate an image using a pith helmet for you: https://sora.com/g/gen_01jqzmab6hfxxtrt3atd0jgpg7
You can just ask for whatever changes you want.
Re: An image of an archeologist adventurer who wears a hat and uses a bullwhip
#460Idk, the models generating what are basically 1:1 copies of the training data from pretty generic descriptions feels like a severe case of overfitting to me. What use is a generational model that just regurgitates the input? I feel like the less advanced generations, maybe even because of their limitations in terms of size, were better at coming up with something that at least feels new. In the end, other than for co…
Tried Flux.dev with the same prompts [0] and it seems actually to be a GPT problem. Could be that in GPT the text encoder understands the prompt better and just generates the implied IP, or could be that a diffusion model is just inherently less prone to overfitting than a multimodal transformer model. [0] https://imgur.com/a/wqrBGRF Image captions are the impled IP, I copied the prompts from the blog post.
DALL-E 3 already uses a model that trained on synthetic data that take the prompt and augments it. This might lead to the overfitting. It could also be, and might be the simpler explanation, that its just looks up the right file from a RAG.