- cross-posted to:
- programming@programming.dev
- hackernews@lemmy.bestiver.se
- cross-posted to:
- programming@programming.dev
- hackernews@lemmy.bestiver.se
cross-posted from: https://feddit.online/c/tech@feddit.online/p/2115447/all-ai-generated-code-should-be-open-source
That i agree on.
Surefire fastest way to destroy ai
Isn’t it already open source? From my understanding nothing created with AI can be copyrighted.
The blog post kind of adresses that point:
However, copyright laws differ across the globe. For instance, countries like the UK grant limited copyright protection to computer-generated works. To make things fair across all jurisdictions, and to honor the open ecosystem that made AI possible, developers and companies should explicitly release AI-generated code under Open Source licenses or dedicate it to the public domain using the Creative Commons Zero (CC0) license.
A legal alternative to the copyright and patent laws we have today is to forgoe the protections, timelines, and limitations those impose, and just… never ever tell anyone what you know. Recipes are a curious example for that, because the technical information of how a dish is made cannot be copyrighted at all, so many companies just keep their formulation under lock and key at all times (Coca Cola and KFC are probably the most famous cases, but they are far from alone in secrecy).
There is a critical flaw for renting AI from the big companies, then trying to keep what you produced a secret: You told them what you asked, and they already know what they answered. For locally generated strings, though - just never tell anyone, and no one will ever know, despite having no copyright protection on any of it.
All
AI-generatedcode should be Open Source.And, as people are starting to find out, all code is “open source”, in that it is incredibly easy to decompile anything. How does that t-shirt go? “Everything is open source if you know assembly.”
Well, existing decompiling tools already knew how to reconstitute assembly into higher-level languages, and LLMs can spend the effort to figure out proper variable/function names to make it more readable.
This is false btw
This will probably be an unpopular opinion but, all AI generated code is already effectively open source.
If person A can generate code that solves a problem merely by explaining the problem to a bot, then what is the point of licensing the code if person B can do the same to generate their own code?
If a zillion people can generate that code, then anyone trying to license the code dooms their project to irrelevance.
The unpopular part: if you imagine for a moment that LLMs can generate useful code and build useful projects, then that dramatically changes the software landscape as we know it. For example, there are software products millions of people are paying a subscription for, which are effectively just CRUD applications behind a nice brand. There’s already some vibe coded bookkeeping / accounting products appearing - its a marketplace land grab.
this line of thinking resonates with my own experiences as a software engineer.
Code has become extremely cheap, it wasnt really a good moat to start with, but now it is basically useless. I suspect software companies will have to compete mainly by investing in marketing, sales, and improving price, and just hope that their clients are apathetic and too entrenched in their products to change to a competitor.
You’re right, and the corollary is kind of terrifying: instead of N people using one tested, maintained dependency, there will be N similar but non identical LLM generated dependencies, each used by ones of people. Talk about code rot.
I think this is a real dilemma: do you contribute to an existing project, going through the motions of creating pull requests and having discussions or do you just make it yourself. I think there’s still value in collaboration and at leat at the moment it’s not like AI can produce a perfectly polished product without some feedback and direction. But I do wonder, if we are heading towards a world where there’s just an AI operating system which can make you whatever you want. Why share something if it is tailor made for you?
I think this is a real dilemma: do you contribute to an existing project, going through the motions of creating pull requests and having discussions or do you just make it yourself.
This might be surprisingly where a coding agent could motivate me. I typically don’t contribute, because in most of the cases I had the urge to get a missing feature into some piece of software, I had to sink countless hours into understanding their code, getting used to their style, bitching about the choices they made, and/or even setting up a dev environment to test the stuff in (which is often extremely badly (i.e. not at all) documented). This then quickly leads to not-invented-here-syndrome. With a coding agent however, almost all of this becomes irrelevant. The agent can figure out how to run it locally. It can figure out within seconds where in the code I need to make adjustments. If I don’t find an explanation for something in the code, I don’t have to search for ages, I can let the agent do the semantic search in a fraction of the time. Now I can focus on what I actually wanted to do: develop the feature, not doing code archeology for hours.
This is something I’ve noticed, so many almost the same packages and projects, with very little actual testing
What’s aggravating is that SaaS companies have been using that model to skirt around GPL. Amazon has Aurora, a MySQL-based relational database, and because they don’t provide a binary, they don’t have to give out the source code. And because they aren’t freely-available binaries, they aren’t subject to decompiling efforts, either.
That’s not to say OSS can’t just provide a binary-version of the same kind of product. But, it does mean that SaaS is going to use that as a layer of IP protection, now more than ever. Terraform (HashiCorp), Redis, and Elasticsearch (Elastic) all are trying to shift to SaaS models to increase profits, stealing the contributions from their OSS counterparts.
Hence AGPL exists, closing that loophole.
I suspect that an LLM can still do a lot of work to figure out how something quacks even if it’s in the cloud.
Moving products to the cloud / SaaS is not a solution, it’s just something to tell shareholders so you’re not doing nothing.
All AI generated code should be blacklisted and shut down.
Whaa whaa
Not a hot take at all
All AI code has been indirectly stolen from open source projects, it only makes sense that the output should be open source as well, see GPL for that.
Ai generated code is useless in that sense. Its like wanting the dilapidated building rather than the structural blocks that its build on.
I can get behind this
Imagine if any work made by AI was free use, public domain, uncopyrightable. Bet that would change the equation for corporations a bit.
Make a movie using AI? Anyone can make copies and distribute them freely.
Make a movie using AI? Anyone can make copies and distribute them freely.
I look forward to that day.
Could you imagine, a worldwide network of people freely sharing movies, music and tv shows for anyone to enjoy.
Maybe eventually we’ll have the technology
Lets call it like river flow or torrent or something.
Make a movie using AI? Anyone can make copies and distribute them freely.
A Chinese outfit tried making a full length film using entirely with AI. It flopped hard.
Betcha it would be everywhere freely as being navigated on the high seas, most probably being MST3K’d.
I’m pretty sure, at least in the USA, there have been rulings that there must be human authorship for something to have copyright protection. This has happened both in the courts and the US Copyright Office has spoken on the matter as well. If you’re interested in one of the court cases I’d suggest looking up Thaler v. Perlmutter.
There was a stipulation though, any human made parts of a work is copyrightable, so in your scenario of copying a movie would probably still be illegal unless the entire movie was created by gen ai, but it would seem that you could probably cut the gen ai bits and release and reuse them without violating any copyright, but I’m also not a lawyer so…
Your right, but it should apply to the whole work. We need to Stop fucking around and put the brakes on them.
I’m not 100% sure how I feel about that take. Part of me agrees, but also part of me doesn’t. Especially with big projects like a movie it could be easy for one person to slip in a tiny bit of ai generated work and now the whole movie loses its copyright? Seems like a bit too much of an overcorrection to me.
I do strongly agree that we need to hit the brakes hard and really work through a lot of the issues gen ai has brought up.
You could say the same thing in the software world. Right now, and it would have nothing to do with AI.
Literally all it would take is for somebody to slip in some open source code and an entire project is suspect.
That gives great incentive for people to stop using it and to heavily restrict its use to only what is needed. If an entire project can be poisoned just like that, larger companies will absolutely hound AI companies to make their output 100% identifiable.
Nuance is the hard part for sure. Im open to ideas.
This is true for at least ai generated images I have heard.
And NEVER sell them, that should be still a crime
Any time you stick a non-commercial clause to a license, it prohibits a ton of stuff because well, you can’t use any of it if doing so makes you money. Like, want to run a website, or make youtube videos, or stream, and you have ads or donations enabled, or patreon or anything like that? If so, you can’t use NC licensed content.
That’s why one important part of public domain is that nobody is also prevented from selling them, so that the content is actually free for anyone to use.
The thing is that, as AI are created from theft, then their output must never be sold.
I’m of the opinion that IP law is fake and made up, and therefore training AI on stuff isn’t theft, and neither is piracy.
It’s also why everything I’ve ever made is licensed either permissively or copyleft.
The problem is thst public domain is worse than copyleft.
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.
Damn that’s a great framing. Will use this talking point going forward.
It’s nearly impossible to prove a negative. But it would be trivial to ask it for code from a GPL project. If it can provide it, it’s in the model.
It’s nearly impossible to prove a negative.
In a general, open-ended situation, sure. In this case, the model developer has all of the training data that was input for training the model. It should take all of half an hour to determine if there is any GPL code in the training data set.
Hi, I am from OpenAI/Alphabet/Meta, I have checked our training data and it contains no GPL code, or copyrighted or pirated media of any kind, it’s all perfectly totally 100% legit. Thank you for your inquiry, hope this solves the issue and have a good day.
Right, so, then we have a paper trail of fraud. Forcing them to make such a statement in writing is useful.
This guy gets it.
Social media users want to see a person walk into the bad guy’s office and punch them in the face, solving the problem. Real effective steps are usually boring and unremarkable like forcing them to make a statement on record.
Exactly, this is how you win court battles that set legal precedents which establish real-world consequences, not as one-offs but every time somebody crosses that line.
Public domain maybe, but not “open source” as in GPL, since the AI training data has issues with an absolute lack of data vetting.
I’ve said for years that all of the “good” or “useful” aspects of LLMs are because they ignore copyright. They’re a glimpse of the creative freedom that IP law stifles.
And pay the rights holders from the training materials.
Also require them to get permission to use the data in the training process before they pay to use it.
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.
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.
Yep. I have loads of PowerShell scripts that will never be used again because they were specific to the company they were written for.
if we weren’t in a game where everyone has to collect as much money as possible, we all could be living in the 2026 that we were promised.
I forget when did world war III start in the Star Trek universe?
It moves around because time travel.













