Except that they are separated. The clean side (model + implementors) doesnt have the original source code. The dirty side is the training room, which does have original source.

The model that you download from huggingface is basically the notebook passed between the two rooms, and as far as im aware, the model doesnt have the actual training data in it, just enough statistics to "hallucinate" something close.

Adobe would have to prove the model that generated this was trained on its stolen code. And Adobe's code contains zero Rust while the Craft suite is 100% rust, so it's certainly not simply regurgitating Adobe's code.

If you learn from others by studying their code, is the organic artisanal code you create also necessarily stolen?

Let Adobe prove AI is trained on their codebase. Fuck Adobe, and fuck the copywrite system.

I think gimp is great software, maybe you just haven't given it enough of a shot to learn it.

To be honest, I think these things will be abandoned very quickly. Therefore I don't think that there will be any competition to the "flagship" FOSS projects anytime soon. However, in the proprietary world, it's money above all else, and I think that we're going to see the quality of proprietary software degrade very quickly. Because why pay engineers to do work if you can have a text generator that does it cheaper and gets there 80% of the way, and then just release it.

I know they are talking about PSDs, they listed it in their test suite. I read it as the test working for all but a few tools, but I guess they could just be talking about coverage.

Either way, it still does and will suffer from all the same issues I listed in my comment.

If I had a dollar for every time I've encountered things in IT that "should be easy to XYZ..." That actually were a nightmare to deal with, I'd damn near be retired already.

And you're 100% correct about Microsoft products. I'll give you 1 guess what company was bragging loudly earlier this year about how much AI-generated code their software engineers are using.

My concerns for LLMs, despite numerous, do not outweight my hatred for Adobe.

I'll just keep using GIMP, thanks

I don't know how anything developed by an LLM can be considered a cleanroom reimplementation. The AI has almost certainly seen the source code of the project it's reimplementing, even of it's proprietary because the developers of said software are likely using the AI too

Yeah we could be a court ruling away from a lot of AI usage being considered illegal. The burden of proof rests on Adobe though. They will have to show examples of their source code being extremely similar I think.

That sounds like a problem for layers of ai companies, not users.

I do not believe that they are doing a proper clean room implementation, but I do think it's possible to do with an LLM.

Let's say for the sake of argument that the model used on the clean room side is a fully self-hosted version of Nemotron. One of the characteristics of Nemotron is that it's not just an open weight model, but from my understanding the training set is open, too. So, at least in theory, Nemotron can be proven to be free of any of Adobe's intellectual property.