That reddit bit is interesting and entertaining (notwithstanding some factual errors), but I find all the discussion of patents to be kind of stupid, because you can't meaningfully talk about them without referencing the actual patent texts. I'm certainly not a patent lawyer, but I've earned a few in my time; one has been broadly cited in its field (which I guess will have to do as far as validation goes, 'cause the company crashed and burned), and another was infringed upon. But time spent talking about patents in vague generalities is time wasted, because you have to be absolutely unambiguous about which specific patent claims are being infringed upon, and how. Wanna talk about Photoshop? Then dig out the documentation and the license and the "About" screens and everything else that lists a patent number, then keep digging through the countless thousands of patents filed by Adobe over the decades that aren't listed and read every single one for a wisp of a reference to any function that might ever conceivably apply to that piece of software, and then evaluate every single one of those claims against what the new implementation does, and how.
Another part of that conversation that doesn't make much sense is the notion of whether or not a generative-AI-created version is (or can be) a "clean room implementation". All this shit's brand new, and I'm pretty sure that with the present US congress absolutely stuck in place, standing on its own dick, there hasn't been any legislation passed (let alone proposed or introduced) regarding these problems. Yes, there have been copyright suits brought, particularly in the music field, but it's still the wild west and there aren't a lot of rules. The problem is that nobody on the outside (of the AI companies, that is) knows a damn thing about what the LLM training hoovered up over the years, so a complainant will have a hard time proving that his copyrighted work was tapped. But when you're a company the size of Adobe (to pursue just this example), that doesn't really matter shit, because all you have to do is push a few mil in the direction of some choice legislators and make sure that if and when a bill appears, it stipulates a reverse-onus, which is to say that instead of Adobe (the plaintiff) having to prove that there was (for example, copyrighted source code) infringement - via the AI and its training - on the part of the slopbros (I love these new terms), it's up to the slopbros to prove there wasn't, which is pretty much impossible, since they're not privy to all of accumulated training to date.
So while it's fun to speculate about the bunfights to come, this is all way premature.
