logoalt Hacker News

All AI-generated code should be Open Source

1116 points • by jcubic • yesterday at 2:46 PM • 599 comments • view on HN

Comments

admiralrohan • yesterday at 4:35 PM

Valid point but this will happen due to market forces rather than moral reasons. Anyone can write code now using AI, so charging for the code itself is getting harder every day, and only the trustworthy people will get the distribution.

The rest of the code will be open source, see how many open Wispr flow alternatives are there already, everyone has same features.

karmugil • today at 1:36 AM

Then all ai models should be opensource

RetroTechie • yesterday at 5:34 PM

"To make things fair across all jurisdictions"

PLEASE stop right there. Nations are sovereign, they decide for themselves if/how far copyrights are enforced. Let's keep it that way. There's enough [800lb gorilla meddling in other's affairs] as it is.

bartleeanderson • yesterday at 7:32 PM

If you type in something like: the history of code and copying previous code versus learning from it and applying that to creative intent behind AI usage for code creation --- you will find that historically that has been a progression from day one. The effort to get good code out of an AI and test/validate it does not minimize the creative intent of the user. Now intent is the operative work. Some intent is bad, some is good. Determining intent is hard for humans as it is for AI. I don't think there will be good answers except for individuals. The courts may rule one way or another. Public opinion in different groups will sway one way or the other. Saying anything is this or that is a slippery slope. The whole premise of this small article seems flawed as a concept and just ignorant of the history. Much like suing someone for a musical phrase of a few notes. There are the same chords/melodies in countless songs and it is not done maliciously. There are only 12 notes, so you figure it out. I think everyone is going crazy about AI on both sides. There are a large group in the middle that are learning how to use it and what it is and is not good for.

freshtake • yesterday at 4:07 PM

Why only code? What about music, movies, images? If AI generated code isn't copyrightable, why should they be?

PeterStuer • yesterday at 4:11 PM

I get the sentiment, but how would you prove a negative? You can get "caught" using AI, but not having been caught does not and can not prove you didn't. Given that 99.999% of projects now use AI to some extend, the logical default assumption is going to be you are probably not in the 0.001%

smartmic • yesterday at 7:23 PM

Related from the FSFE, with a bit more of legal background: https://news.ycombinator.com/item?id=50489088

AshleysBrain • yesterday at 3:58 PM

Related question I just thought of reading this: if you are an employee at a company that you know 100% AI generates its code, and AI-generated code is not copyrightable... is that employee legally allowed to leave the company taking a full copy of the source code, and set it up as their own business?

➕ show 3 replies
v4vvdq • yesterday at 7:47 PM

Not the AI's source code output should be open source but the prompt should be. The prompt is the real source and can be used in the future with newer LLM models.

smilingPanda • yesterday at 8:11 PM

I think under the code lives ideas. If somebody figure out some nice and cool thing, and coded it with llm - it's still it's idea.

ganelonhb • yesterday at 7:04 PM

I disagree, however companies who decide to work on proprietary code need to be doing comprehensive license similarity scans on their codebase or risk licensing issues

yitchelle • yesterday at 4:52 PM

Open source is not always open source as we all know there are so many different flavours of it. How do you make that distinction when using AI to generate your code?

This is a debate that we need to have before it gets event more muddier.

ZiiS • yesterday at 8:16 PM

Probably but zero chance, so back in the real world how do we keep people motivated to maintain open source software going forward?

mococa • yesterday at 10:11 PM

To be honest, AI generated code must never have a license or a very permissive license

semitones • yesterday at 9:40 PM

> The second reason is legal: purely AI-generated code is not written by a human, meaning it should not be copyrightable.

Humans are reviewing it (mostly), and choosing to merge/deploy it. Imagine you had the million monkeys on typewriter, and a human kept reading samples until they chose one. The human is in the loop, so I don't think this argument applies.

Cort3z • yesterday at 6:20 PM

I agree, but also, any code that a company can generate and use, an open source solution can generate and release, effectively eating the corporation’s lunch in many cases.

mahboi • yesterday at 7:38 PM

There's a decent change that AI makes closed-source software not really a thing anyway. So it'll either be open or unreleased.

pbhjpbhj • yesterday at 4:41 PM

No, copyright protects works made by humans.

AI generated code should be (is?) public domain.

superkickstart • yesterday at 3:29 PM

All 100% ai generated assets, art, music, text should be free and open source too.

➕ show 1 reply
iamflimflam1 • yesterday at 4:59 PM

Somewhat amusingly, I just posted this... https://news.ycombinator.com/item?id=50508364

lhk931122 • today at 1:22 AM

The web is already full of AI-generated slop like posts, images and videos, and if we just say the code has to be published without any plan, the web will fill up with slop and it will get hard to tell what is worth referencing and what is not. It seems better for everyone to just keep their own work private and only publish what they think is worth publishing.

csmoak • yesterday at 4:03 PM

Fully AI output from a single prompt is not copyrightable (in the US). Is fully AI output from a sequence of prompts not copyrightable? If it is, at what point does this change?

➕ show 1 reply
gnarlouse • yesterday at 9:35 PM

@jankiewicz: how are people expected to run a business under this model?

low_tech_punk • yesterday at 6:26 PM

What is user changes 1 line? 2 lines, 3 lines...? The concept is well intented but I doubt you can enforce it.

resident_ • yesterday at 5:14 PM

I use AI to develop apps for manufacturing but I still have to think and point AI how things will work together. If I simply tell AI to develop something complex most of the time it will do a really bad job doing so. I do have a coding background but with AI I can focus on other things and use the time I’d spend coding to work on other problems.

simultsop • yesterday at 10:05 PM

Open or Closed. If it is not human engineered it will just amount to the pile of generated slop. If the model generated that version it will be able to do it infinitely. I fail to see what would it fix. It's like if we ask everyone to display their socks or something trivial. If people keep something cheap private, I am puzzled with the others paying for it. We already know who trained models at first place, now what does this ask bring from consumers, I don't believe it will pay any credits to anyone.

On the other hand, I can't imagine doing what I do with AI today without it in this velocity.

The incremental complexity of stacks was a torture to engineers, up until recently. Now we can steer plenty of AI resources to where customer wants to invest. Where nowadays everyone is building ecosystems, platforms. Not a tiny integration.

rib3ye • yesterday at 7:36 PM

Couldn't an org still constrain distribution (keep the repo private forever)?

neya • yesterday at 6:46 PM

> purely AI-generated code is not written by a human, meaning it should not be copyrightable

This is a very poor definition. I use Phoenix's scaffolding all the time. So, running

    mix phx.gen.html Blog Post posts title:string content:text 

means my code is no longer copyrightable since it was written by a machine?
wolttam • yesterday at 3:27 PM

There is so little value to most code produced with models these days I just don’t see the point.

The real value ends up being found in software where there is a person behind it continuously refining it into something others will look at and say “Someone clearly paid attention to this.” Models can’t (yet) reproduce that because they can’t/don’t go interacting with users or asking bigger picture questions.

Whether someone is producing that software with a model or not doesn’t change much about the ownership picture to me: it’s the person that owns the project, and the LLM is a tool.

Once again, it all comes down to it being a tool and how it’s used, and there are no useful blanket statements that can be made about it (or how the tool’s output should be classified, legally or otherwise)

temporallobe • yesterday at 5:35 PM

Even code generated for government organizations with various levels of secrecy? Not gonna fly.

cluckindan • yesterday at 5:17 PM

The code may be non-copyrightable, but the business logic may be proprietary and a trade secret.

frigg • yesterday at 4:10 PM

Does it matter at this point when you can generate the same code easily with the same AI model?

➕ show 1 reply
kleiba2 • yesterday at 6:30 PM

Could LLMs be considered derivative work in the sense of the GNU GPL?

moneybunny • yesterday at 3:37 PM

If you put your code online just assume it's gonna get stolen and you'll never see a penny

➕ show 1 reply
peter_d_sherman • today at 1:40 AM

A great legal question!

That is,

What is the exact legal status of any AI generated output?

(Of which but one of the many subquestions is "can the AI generated output be copyrighted if an actual human didn't produce it, and copyrights are for human produced work only?")

Now, I'm sure I'll probably get several "this is how the Law works" responses to this...

But I'm not really looking for such responses; that's because I've studied a lot of Law, and I know that Laws can define things in various ways, and Courts can decide things in different ways...

Phrased another way, I'm not looking for a "solution" for the above "problem"...

The question, on the other hand, in its purest form, is the root of a larger philosophical discussion (even before we get to the legal one (i.e., Legal Discussions Vs. Philosophical Discussions -- what's "legal" Vs. "what's the right thing to do?") -- are the kind of questions which I seek!

Anyway, a very thought-provoking article!

I'm not saying I necessarily agree with the premise of it, I'm not saying I necessarily disagree with the premise of it...

What I'm saying is that it might make an excellent root reading for further face-to-face human discussion, further questioning, further philosophical discussion, and the like...

Anyway, a very thought-provoking article!

dgggh • yesterday at 4:52 PM

As the cost of producing software approaching zero. The legal side go down with it.

earth2mars • yesterday at 5:31 PM

A pre-cursor. All AI models should be open weights/source to begin with

PLenz • yesterday at 4:51 PM

AIs don't have copyright at all. All AI generated code is public domain

trescenzi • yesterday at 6:01 PM

All models should also be open source and socially owned. All of this output is the culmination of human knowledge and should be treated as such. I get this is about copyright but I don’t think people are thinking big enough with this stuff.

the_arun • yesterday at 4:36 PM

If AI services that created the code were free, I agree with the sentiment.

jonathanstrange • yesterday at 3:56 PM

This article makes no sense. None of the arguments are sound and conclusive. First argument: Non sequitur. Programmers were also trained by open source software and from stack overflow and stack exchange. That's not a sound reason for their programs to be open source. Second argument: AI-generated content is not copyrightable. If that's the case, then it cannot be open source since open source licenses rely on copyright for their enforcement. The author confuses public domain with open source. Public domain and open source are two completely different types of categories with different legal implications. Third argument: This contradicts the second argument. Now programmers using AI should publish their code under an open source license or under CC0. That's a demand. Not everybody will do that and no argument is presented why they should do so. Fourth point: Not all code is a hundred percent human written and a hundred percent AI written. The author concludes from this problem that "Code like this, where humans put little effort, except writing prompts, should definitely be Open Source." There is no argument here at all. It's another personal demand without justification.

At least, the author should try to back up their demands with further justifications and make them less contradictory.

➕ show 1 reply
partsch • yesterday at 7:53 PM

No, but all models should be open weight.

0x20cowboy • yesterday at 6:48 PM

I kind of agree, but like most software engineering takes I think this misses the point of software as a whole.

What the software _does_ is almost always more important than how it does it (save for some amazing new technique which is very rare).

There are only so many ways to write a binary search or an octree, but the overall idea behind, say, doom is unique.

Open sourcing big blocks of code generated using standard techniques isn’t that useful (since you can just generate the same ones yourself), but the concept behind the application is the important bit (which of course, is now also easy to just generate).

If anyone can just generate their own version, who cares if it’s open source?

Pannoniae • yesterday at 3:24 PM

This post is so close to getting it ;)

Looks like the author wants to abolish copyright in a very round-about way. They are saying that AI-generated code should be released under CC0 due to complexities around the status of software copyright in various jurisdictions.

But if the concept of copyright was abolished as a whole, then licences such as GPL would also lose their power. Some people say that's a negative thing, but I do not necessarily think so, because it achieves the stated goal of the GPL, software freedom (i.e. you don't need to be granted a licence to copy and modify things)

➕ show 3 replies
delichon • yesterday at 3:23 PM

> First of all, AI models were trained on a massive corpus of Open Source code written since RMS announced the GNU Project and the Free Software Movement.

I too have trained on the massive corpus of open source code. Should everything I code also be open source? Or only if I leverage an artificial neural network as a prosthesis to enhance the capabilities of my natural neural network?

➕ show 6 replies
hauntingseaweed • yesterday at 6:30 PM

I completely agree with the OP. What irks me the most is that Amodei has trained his model on someone's else IP and now he wants to resell this to people and has the nerve to complain about distillation. Distillation should be god given right. I guess that is why they took down Jensen.

fathermarz • yesterday at 7:11 PM

What problem does this solve exactly? I think when most code becomes generated it’s all from the same hivemind of the human collective. I see arguments that code is not the hard part of development, so why are we treating it so sanctimoniously now?

angry_octet • yesterday at 9:02 PM

There is nothing legally supporting this, as LLMs are a tool used in our creation of products. However, it would be hilarious if their code was ruled to be GPL v3.

🔗 View 50 more comments