The real rub will be the first court precedent on whether GPTs infringe on source data IP.
Could see it going either way: fundamentally transformative or not.
I'm not sure how well it will go down in court fighting this... since we agreed to it. But the more interesting question will be is the result a complete "you loose" and GitHub walks away, or if they are forced to take actions in order to defend the copyright of users producing content... a "Code Id" type system that warns you if the code your uploading is too similar to someone else's in order to allow you to use the fun new AI tools to make code and pay GitHub, but also simultaneously defend users legal intellectual property rights.
The relevant section is this:
GitHub Terms of Service: Section D, Sub-Section 3 - https://docs.github.com/en/site-policy/github-terms/github-t...
The phrase relevant to your point is "If you're posting anything you did not create yourself or do not own the rights to, you agree that you are responsible for any Content you post; that you will only submit Content that you have the right to post; and that you will fully comply with any third party licenses relating to Content you post."
It's fair to interpret that as GitHub are not going to be copyright police. The bit at the end where I suggest a "code Id" is more of a thought experiment as to how they could continue to offer the service while complying with a potential adverse ruling that doesn't ascribe blame on them or the service since theres another section of the ToS that I, with my "knows slightly more about law than average but absolutely not a lawyer" hat firmly on, feel will be how GitHub's legal team at least try to make short work of the lawsuit, their success with this tactic is a matter for the Courts, and I'd love better legal scholars to weigh in.
GitHub Terms of Service: Section D, Sub-Section 4 - https://docs.github.com/en/site-policy/github-terms/github-t...
Which reads thus:
"We need the legal right to do things like host Your Content, publish it, and share it. You grant us and our legal successors the right to store, archive, parse, and display Your Content, and make incidental copies, as necessary to provide the Service, including improving the Service over time. This license includes the right to do things like copy it to our database and make backups; show it to you and other users; parse it into a search index or otherwise analyze it on our servers; share it with other users; and perform it, in case Your Content is something like music or video.
This license does not grant GitHub the right to sell Your Content. It also does not grant GitHub the right to otherwise distribute or use Your Content outside of our provision of the Service, except that as part of the right to archive Your Content, GitHub may permit our partners to store and archive Your Content in public repositories in connection with the GitHub Arctic Code Vault and GitHub Archive Program."
For me the key quote being "including improving the Service over time. This license includes the right to do things like copy it to our database and make backups; show it to you and other users; parse it into a search index or otherwise analyze it on our servers; share it with other users;". Now theres some legal arguing to be done about if charging for the AI constitutes an infringement on the second paragraph which opens with "This license does not grant GitHub the right to sell Your Content. It also does not grant GitHub the right to otherwise distribute or use Your Content outside of our provision of the Service" but thats a very different argument to what I see a lot of people making. People are arguing (generally speaking) from the standpoint of "its not right, this violates my rights as the author having selected this license for my code/commits and published it under that licensed for others to share" ... not "i didn't agree to GitHub selling my content and this constitutes violation of GitHub Terms of Service: Section D, Sub-Section 4 where they told me that they would not sell my content".
But broadly speaking, unless the argument shifts to GitHub Terms of Service: Section D, Sub-Section 4 and classifying this as an unapproved sale of the users content, then I don't see how GitHub are not well within their rights to have trained the AI model and offered it as a service. We by agreeing to the ToS agreed to GitHub Terms of Service: Section D, Sub-Section 3 where we promise to only post code we have the rights to post and that we the user will comply with the legal complexities of third party licenses and basically are responsible for not posting stuff to GitHub for which we cant grant GitHub the requested legal rights, which when combined together means that we gave them permission to use our code and commits, regardless of any license files we may have put in the repos, to train the AI model. We can definitely argue derivation and what justifies a sale, and I'd be inclined to say they may actually have breached that term, but no one I've read is talking about that, its all about copyright infringement for AI generated code and moral rights with respect to using the code to train the model, not a clear cut contractual breach of the Terms of Service that GitHub may or may not have perpetrated on us as the other party agreeing to be bound by the contract.
GitHub ToS is written by GitHub, it's not a contract in the sense that no consideration has been given to the other party and as such it isn't legally binding on that other party, but regular law, such as copyright law, still applies to GitHub.
Even GPL doesn't (yet) include a clause saying the code can't be used to train AI unless the AI itself is open source
That's not what they allow for, and copyright being a 'right' it allows you to pass those rights on to others and to retain some for yourself. If not explicitly passed on the right still rests with the original author, plenty of precedent for that.
Therefore isn't following the terms of the copyright grant, ergo doesn't have a license for use, ergo is violating copyright.
Now what does that look like when I take 100 different open source licenses, including MIT, put them in a GPT blender, and then productize my output without following any of the licenses?
... makes you think there might be a legal component to why OpenAI switched to a SaaS model. Although believe they'd still be in hot water over any AGPL et al. code.
A: Common ToS to say that a product's owner obtains a license to user content for purposes of providing that user the product service.
B: Somewhat common ToS to extend that to providing the product service to third party users (i.e. use your content for other users of the service), but depends on business model (e.g. most social* businesses).
C: A lot less common ToS to obtain a right to distribute user content in derived products.
A number of sites have gotten into hot water with their userbase over trying to update their ToS from B to C. From memory... Adobe Cloud, DeviantArt, maybe some others?
Typically this gets flak in creative communities, given that it is many people's business, and they're more concerned about distribution rights than your average coder.
At its base, OpenAI/Microsoft/etc. will eventually run into the exact same issues that bedeviled the Linux kernel in the 1990s, except with a much thornier IP ownership question (given the greater number of parties).
1. Huge company.
2. Impossible to prove that a weight of -0.7 in a neural net means they used your code.
3. The code spat out by the bot isn't your code.
It does unless you can afford to sue Microsoft.
You can't opt-out of those terms, regardless of your license, just as you can't opt out of Facebooks terms which give them the right to use your content for their business or marketing purposes, even if the various chain posts that have spread there might claim otherwise.
If the code is GPL licensed, you can't relicense it under a non-FOSS license like you're talking about
> Don't they then have to remove that code?
Yes, if code gets uploaded whose license is incompatible with Microsoft's terms, they probably do have to remove it.
> Now imagine this: I write some code, license it, but don't publish it yet. It's licensed. Then I upload it to github. Does the license supersede the ToS? Doesn't github have to remove the code as a ToS violation
Again, yes, they probably do, and they also probably have an obligation to clean their training data of it. However, if you're the copyright holder of the code and you agree to their ToS before uploading, they might make the case that you agreeing to their ToS does grant them the license to use it in training data.
So then Github is breaking the law a significant portion of the time then at the very least?
Facebook wouldn't be able to allow users to upload photos.
Hacker news wouldn't allow me to post this comment (someone else could own the copyright, right?)
You are violating the law, probably. The ToS would say something like "I hereby declare that I hve the right to agree over the software to submit it under the ToS".
However, the reality is that this is all extremely muddy, far from proving that software A has copied some code from software B where you can just compare the source code. There are too many muddy steps, and you can bet that Microsoft will just get away with it.
It remains to be seen if they have a way to then clean their training data of the influence.
It would be the same situation if someone uploaded any other proprietary code.
Copyright isn't 'opt in'.
Before 1989, copyright protection was opt-in in the US.
If the github ToS supercedes the author's own licence, then I guess the uploader is effectively relicensing the code without the author's permission. That would mean the author has cause for action against the uploader, but not against github.
I personally dislike git; I find it too complicated, because of features I don't need. Microsoft has always disliked FOSS anf GPL, and I suspect that Copilot is a deliberate effort to undermine it.
But in what way does reading a copyrighted work and then producing a mass of numbers as a result infringe copyright?
E.g. the courts take a dim view of any attempt to create a machine (in the abstract sense) that takes in copywritten works and churns out similar-but-uncopywritten works
Edit: I googled "fair use copyright US" and have now decided that US copyright law is stupid.
Don't be like that. Fair use is what allowed VCRs to continue existing, what allows Google Images and Books to exist, what allows the development of emulators... I could go on.
How long exactly did you study the issue? It is very complex.
Fair use helps artists, journalists, and (indirectly) the general public. It prevents censorship of critical or opposing views, among a lot of other uses that are beneficial to society (see sibling comment). US copyright law is backwards in a lot of ways but this isn't one of them.
Fair use is a statutory right (codifying what courts had 0reviously found to be an aspect of Constitutional free expression rights limiting the copyright power) that limits the exclusive rights of copyright owners. It can't be reduced in scope by license terms, because it deals with what the owner has no right to control in the first place.
(You may be confusing “fair use” with “implied license”, and, yes, explicit license terms would be a powerful argument against an implied license argument.)
My impression, they claim the fair use only because there is no other ground for ML training on GPL code. The license only describes other uses of the code, so the first question is on what grounds Open AI uses the code at all. And the only thing OpenAI comes with is the fair use.
ML training does not fall under free expression or similar basic rights, I think. Only because it looks too restrictive to forbid scanning the code which is publicly available for people to read, and the license itself does not mention such use, the fair use may be considered as some kind of justification.
I, of course, can be mistaken. Especially that I didn't study this subject deeply.
Also, for authors who are against the ML training on their GPL code, it's better to prove it violates even the current GPL version, rather than just introduce a modification to the license. Because new license will only cover new code, while old versions are already available under the old license.
On the other hand, such an amendment to the license will not be harmful, even if later proven to be useless. And it will clearly state the copyright holder's position regarding the ML training, instead of some subtle reasoning we should apply currently.
Some people may want an opposite amendment to their license - ML training does not produce a "work based on the Program" and can be freely performed by anyone.
But: NOOOO.
In order to close off this possibility, which would restrict Copilot revenue, they instead would roll out a single undifferentiated product and with lots of "gee whiz!" and associated hooplah, and be sure to offer it for free for a while to suck everyone in and head off criticism.
"If you train an AI on this code, you must release the source code and generated neural net of that AI as open source" or something to that effect.
It won't stop it, but it will slow it down, and it seems like the right T&Cs to put on training against GPL code because it gives an advantage to open source AIs, however minor.