Hacker News Reader

> Yes, I'm talking explicitly about feelings. I know my feelings don't impact the legalities of a license.

They don't impact the current legality of a licence, but it will affect future ones.

GPL/BSD/Apache/proprietary, they are all picked for ideological concerns which all stem from feelings. It is good to discuss these things, and it is good to recognise that these are emotionally driven.

> They don't impact the current legality of a license, but it will affect future ones.

Don't they? Even the most liberal licenses require that you at least keep the license and attributions. Which are exactly the parts that AI systems remove. I would have no problem with an AI system trained on GPL code if the output was still covered by the GPL.

The MIT No Attribution licence https://choosealicense.com/licenses/mit-0/ doesn't require an attribution.

It's the licence I choose for new projects.

That's very generous of you. Don't get angry though if I go ahead and re-publish all your code with your name replaced by mine. That's exactly what you are allowing me to do with such a license.

That would be a copyright violation, not a licensing issue. GP is giving you a _license_ to use his code without having to say "GP wrote this", he is NOT giving up his rights to the code..

edit: to clarify, he's allowing you to _use_ it however you like, including making a derivative work, including it wholesale, etc. However, claiming authorship of the original code would still run afoul of the original copyright.

edit2: oh - if you mean relicensing the code - that is allowed.

Whether or not copyright applies at all to model training is an entirely open question, and where rulings have come down, it's likely closer to these situations being fair use (e.g. the Google Book's case, which was ruled transformative and not a direct replacement for the works in question).

The reality is, these models don't copy or distribute anything directly, which makes applying copyright a bit of a stretch. Many people feel like it is a use that should have some sort of IP law applying to it, which is why I think there's some chance that courts or legislators will decide to screw the letter of existing law and just wedge new interpretations in, but it's not super simple: they'd have to thread the needle and not make things like search illegal, and that's tricky. Besides that, these models are out there, they're useful, and if they're ruled infringing they'll just be distributed illegally anyways.

I don't envy the people who will have to decide these cases, I suspect what's better for the world overall is to leave the law as-is and clarify that fair use holds (nobody will stop publishing content or code just because AI is slurping it up, a few weirdos like the article author excepted), but there are going to be a lot of pissed off people either way...

Would I be able to train an AI only using microsoft's leaked windows code to write a windows clone with no copyright (since it comes from an AI) and be safe from legal repercussions because it was trained on fair use code I just happened to find online?

If they rule that it's ok to do that, I might be ok with AI being ruled as fair use.

Is it even fair use under the law to consume that Windows source code?

People who train AI models claim that anything they can find on the internet is fair use.

It's the whole point of the discussion… is it really?

And if it's not fair use to train on windows source code because of copyright… doesn't that same copyright law cover everything else as well?

I think there's a reasonable distinction to make between "you can train AI models on any code that you are legally allowed to have and read" and "you can train AI models on any code that you are able to feed into it, regardless of whether you have permission to possess/read it".

You're not legally allowed to have and read it if you knowingly violate its license terms.

Exactly. I'm betting if you asked GPT to create a windows clone, for sure MS would not let you distribute. This will go like every other law/license, big corp can sue little guy into the ground. When big corp uses your code it will be 'thats just a model generated code not yours'. But in other direction, if little guys creates windows clone, 'sorry, its off to jail for you maytee'.

Even if it's the opposite direction, big guy losing and small guy coming out ahead, it's still drama.

Just like Covid introduced epidemiological terms to the general public, this issue can introduce design choices around licensing, copyright and watermarking to more people.

I assume there is a group of researchers building tools to provide fine-grained historical views into AI output. And yes, for billions of parameters trained on billions of documents, linking every letter to a source document is a UX nightmare.

But what a cool problem. That's the interesting part. Yeah, something like TileBars[1] or Seesoft[1] seems like the right tool. But maybe keeping it all text with some graphical marker of authenticity is the better choice.

So many cool problems. But, that authenticity marker is the hard sell. Can reasoned discussions with others be enough to introduce that, or is drama required?

https://people.ischool.berkeley.edu/~hearst/irbook/10/node7....

You misunderstood. Their emotions do not impact current licences, but their choice of licence is an emotional act.

There are plenty of arguments to choose a licence, and as the world changes we will evolve. See how the GPL itself evolved to handle TiVo. These things arent static.

I personally agree that GPL trained code should produce GPL code, I see a distiction between teaching people and teaching computers but that isn't my call.

They are emotionally driven, but that is not all there is too it. They are also driven by an insight, that most people/organisations do not give back to the community, unless obligated to do so. It is about advancement of society and goods everyone can use, if only they give back to society, when it is their turn to share.

This is exactly what code laundering ANNs circumvent and that might open up a dystopian future for all of us, not only us code monkeys, but society in general.

>they are all picked for ideological concerns which all stem from feelings

Or, perhaps more commonly, they were picked for business model reasons or (related) because building a community that includes commercial interests tends to favor more permissive licensing.

Yes. People here seem to be forgetting that Open Source was a community driven ideal first. The License came later as "protection". Corporations were stealing code and there was no recourse. The variety of open source licenses were created to provide a framework for the community, to fight off stealing, to keep it open. So GPT is very much 'laundering' the code just like criminals 'launder' money.

I agree that AI usage of code is somewhat murky with current licenses,which obviously don't mention it either way.

Free software has a principle of "freedom to run, to do whatever you wish" (freedom 0), so arguably has said that training AI is OK. (We could quibble over the word Run, but the Gnu.org,and RMS clearly say "freedom 0 does not restrict how you use it."

GPL code can be used by the military to develop nuclear weapons. Given that the is a guiding principle of the FSF its hard to argue that the current usage is not OK.

I have no problem with Copilot being trained on AGPL code and the getting released with a AGPL compatible license. Free to do whatever they want with it.

The problem is Copilot training on source code and then discarding any restrictions of the licenses. Maybe it is legal right now but I'm sure this case will find it's way into open source licenses pretty soon.

Do we know if CoPilot X was trained on AGPL, not just GPL?

Additionally I'm not sure if AGPL does anything.

I suspect the ethics and such of licensing when large fractions of work are training AI and using AI need to be worked out rather than getting mad at any individual.

What does copy left look like for AI?

There is indeed a problem of transparency right now. Companies afaik did not release the complete training data set. Might even be intentionally, because they do not want to risk, that they trained it on stuff they should not have had, without building in license and attribution into the output of their models. Or it might be, that they know that to be a fact.

I can only hope, that lawmakers hurry to catch up with reality and impose transparency obligations for AI models.

Even if usage is legal right now, the other obligations of the license need to be adhered to as well. Can't just pick or choose one tiny aspect of FSF philosophy and run with that. AGPL is clearly about sharing and spreading free/libre software as well.

The problems are attribution and licensing. Not use.

I largely agree with you, but I think there is one question that hasn't been addressed yet: Are the weights learned by an LLM a derivative work?

When a person learns from GPL code this question doesn't arise. The state of a person's brain is outside of copyright. But is the state of an LLM also outside of copyright or outside of the terms covered by the GPL? I'm not sure.

An LLM can not only emit source code derived from code published under the GPL, it can also potentially execute a program and could therefore be considered object code.

This isn't necessarily a problem as long as the model isn't distributed and does not include any AGPL code.

> the state of a person's brain is outside of copyright

It clearly isn’t. Which is why clean-room reverse engineering always requires at least two people. Or why a musician that accidentally recreates a chord progression they heard years ago but don't remember the source might still get sued.

No, you're missing the very distinction I'm trying to highlight.

When I read and remember some text, possibly also learning from it, I'm not making a copy and I'm not creating a derivative work. The state of my brain is outside of copyright. Only at the point where I create a new representation based on what I have read I may be violating someone's copyrights.

But is it the same for an AI? Is the act of reading, remembering and learning (i.e. adjusting model weights) not in itself tantamount to creating a derivative work?

> The state of my brain is outside of copyright.

Is it actually? If we could fully pull out the state of your brain, and understand that you stored a copy of a copyrighted work, I think you could be on the hook for licensing it, paying fees every time you remember the work as a performance of it.

The state of your brain is moot wrt copyright as you cannot distribute your brain.

Copyright is to the exclusive right to make copies, it is the exclusive right of distributing them.

As a simple example reading a book aloud in your home or singing in the shower is not copyright infringement; not even if you record it.

If you sell tickets to these performances or stream them on twitch it becomes copyright infringing.

Similarly it cannot be in violation of copyright for GitHub to train copilot on any random code they can legally access. It can be in violation to sell access to the model trained in this way.

Fully agree.

This may seem a bit nitpicky and philosophical, but anyway: these feelings you mention are about things, and these things the feelings come from are what is most important. Feelings are never standalone, if they are they are just moods which are so personal its hard to have a conversation over.

Let's call 'the things' values. I'd say feelings are perceptions of values, and as such they invariably have a conceptual element to them. And exactly that conceptual aspect makes them suitable for conversation and sometimes even debate, insofar as they can be incorrect. We can acknowledge the subjective, emotive aspect of feelings as highly and inalienably personal, respect the individual opinion behind them and contest the implicit truth-claims all at the same time.