All AI-generated code should be Open Source
Which is fine, no one cares, they're not redistributing or selling it so licensing probably isn’t needed or important. (IDK not a lawyer).
But if someone does want to sell or redistribute, then any model with, say, GPL input should (I think) be passing on the GPL terms with the redistribution of the derived work.
It does get tricky. Cause I can read open source code and learn to code. Then code something and sell it. I just can’t use the exact code word for word. And if you forced all companies that use AI to open source their code, there would no longer be a reason for those companies to exist. That doesn’t work out well either. Maybe make them contribut to an open source foundation or something.
Loads of open - or at least open to club - standards exist that people work to, and make a living off implementing well.
Companies can exist to efficiently implement them, capture economies of scale, pool supply chain risks, develop compatibility and interfaces , create an environment where teams can cooperate or specialise and reap the benefits thereof, they can develop QR processes building customer confidence through reputation, develop customer relationships so as to clearly identify requirements; and so on.
Also just being an organisation that will stand by its work, and exists as a legal entity to be sued if it fails, is a benefit to some customers - see professional service for example. The insurance and liability and building customer/client trust one of the main parts of many businesses.
So, I think there are plenty of reasons for companies to exist without owning IP on code (that they redistribute).
They may like IP on code because i think they, and investors see it as high margin - whether they're right or not is another thing; but high margin activity should only exist temporarily to be bid down by competition sooner rather than later (in a well functioning economy).
But even so i think the majority of those processes, knowledge and code would not be for redistribution anyway - and a lot of the really good stuff is often tacit knowledge anyway.
It'd all be totally unenforcabe anyway; I think this just a moral flight of fantasy, but if other people worked in the open in faith that any derived work would also be open, then a moral user (aka the unicorns) should respect that.
Ok, I’m on board with what you are saying. Profession services would still be viable for those reasons.
This is exactly how it should work. Any code copied from GNU GPL code should legally also be GNU GPL, no ifs, ands or buts.
Oh, your training data set contains GPL code? Guess what, all the output from that model is now GPL. It’s your burden to prove that your training data does not contain any GPL code.
LLMs obviously don’t copy code. That’s not how they work.
What you’re suggesting is sort of the same as claiming that if a developer ever read GPL code, all their code is now GPL code.
(I think I’m in favor of this fwiw)
Maybe going against the wind here, but this doesn’t make sense to me for generated ideas and code to be open source. If I have a unique recipe that made something valuable, I should be able to have that power of decision over my idea. What I think would be a must for this to be realistic is that all LLMs that trained on public data be, by law, public domain. So a company would still own how they did the training so there still would be something where that expertise is rewarded.
I agree, it’s more about the economics of AI use itself rather than the end product that’s concerning to me.
I think you would be suprised how much of the code that is written by AI is solving the same problems every company solves, and is totally useless outside the company. Very little of it is really novel or valuable. And that is the code not being written by AI.