Comic Strips

I just hate that so many people are refusing any AI tool because it is AI... in my opinion there are places, where these tools are really useful.

Well, we can agree to disagree. Lol! Seriously, I honestly don't have much charity for what is being called "generative AI" lately. Wouldn't be caught dead using it. (Hell, I'm a mod of the "FuckAI" community here on Lemmy. Which isn't actually saying that much because there are a lot of mods in that community. Heh. Anyway.)

But! To each their own.

So lwt me repeat, codecomic sound awesome and I only wrote my comment to start a discussion.

Thanks! I'm glad to see some interest in it!

I have one question, would it be okay to create codecomic source code with LLM?

Well, as someone who is vehimently anti-AI, I don't really think it's "okay" to use LLMs or Stable Diffusion or whatever at all. But I wouldn't say it's any less ok to use an LLM to generate codecomic source code than to generate... I dunno... C++ source code or OpenSCAD source code or whatever. (Last I heard, the chat bots were so abysmally bad at OpenSCAD as to be completely infeasible to shoehorn into that use case. Not sure if anything's changed about that. Similarly, I'd think with codecomic being as obscure as it is with very few example code snippets out there to train on, I'd be really surprised if any LLMs out there would be even a fraction as good any time soon at codecomic even as they've ever been at OpenSCAD.)

(This is probably a good point to say IANAL and none of what I say is legal advice.)

Beyond that, I purposefully licensed codecomic under the GPLv3 on the premise that I'm not ok with anyone else distributing any portion of codecomic (or anything I Open Source, really) without ensuring the recipient has the corresponding source code. (Also, codecomic's source code has a "terser" binary form that can save space, so there is a rough equivalent to the difference between "compiled" and "source" code when speaking of codecomic code. Plus, codecomic can be used as a Go library instead of using the codecomic DSL and Go is a compiled language.) Given that basically the only codecomic source code publicly available right now is in my repository and covered by the GPLv3, any codecomic source code any LLM would be trained on any time soon would be licensed under the GPLv3. And any codecomic source code an LLM output would pretty much have to be so closely copied directly from that that I'd say it would ethically be well within my rights to demand attribution and compliance with the "source code provision" of the GPLv3. (Honestly, I'm not likely to spend money on lawyers to force the issue really. I might send a cease and desist or something if I found out that was being done, though. But you didn't ask whether one could get away with it. Just whether it's "okay".)

I know in some jurisdictions, courts have basically ruled that the copyright owners of the training data don't own the copyright on the output of LLMs trained on their work. (I think some courts have ruled that the human running/using the LLM doesn't own the copyright on the LLM's output either because to be covered by copyright requires specifically a human to exercise some amount of creativity in the creation of the work. One might be able to argue that "prompt engineering" qualifies as "creative". I hope that argument doesn't fly, but who knows what the courts will eventually decide.) But in those cases, the LLMs would be trained on data from such a diverse set of sources, under such a diverse set of different licenses, and from such a diverse set of authors that it can't really be tracked down "which works" are being copied. But in the case of codecomic, unless/until wider-ish adoption happens, I'm thinking it would be a harder sell to try to claim that LLM-generated codecomic source code isn't a derivative work of codecomic itself.

But again, if lots of people started publishing codecomic source code under licenses that don't have a "source code provision" or anything roughly equivalent, I don't think it would be any worse to use an LLM to generate codecomic source code than to generate source code in any other particular language. Hopefully that answers the question. Heh.

And one final note on AI tools. Some of them are really complex. I don’t use them much, but some of them allow user to edit only parts of the image with a prompt, or create an image based on user sketch, etc.

Yeah, I wouldn't say any of those use cases really make LLMs any more similar to codecomic, though. There's still a big difference between the two approaches to producing an image or webcomic or whatever.