5 ms·

GitHub Copilot Suit’s Copyright Claims Are Likely a Dud

3 points by sam345 4y ago

belorn 4y ago
> The plaintiffs allege that Codex and Copilot violates the open-source programmers’ copyrights because Copilot will sometimes regurgitate code it was trained on This misunderstanding seems to be in the core of the discussion. One side thinking that copilot can write novel and unique new code using the mechanics of the programming language. The other side do not agree, instead seeing Copilot as simply an complex algorithm that regurgitate portions of the training data depending on context. The article do later go into a bit of discussion if those portions of training data can be considered copyrightable, but there are precedent. If you take a copyrighted work and split it into many small pieces, each much smaller than any copyrightable work, they can still be considered protected. Bittorent is one example, but there has also been examples where paintings has been cut into physical pieces and made into new creative work. In both people have argued that the individual portions aren't copyrightable and lost that argument in court. > Proving that all or even some of the code the purported class members wrote are copyrightable is not possible to do on a class-wide basis. It would require a judge or jury to opine on an instance-by-instance basis whether a given class members’ purportedly copyrighted code was original enough to qualify for a copyright. The pirate bay founders made a similar argument. Their lawyers argue that the court would had to first opine on an instance-by-instance basis for each work that was claimed to be pirated. The judges however said that such instance-by-instance work isn't necessary unless there was doubted that copyrighted work had been copied through the site. If Copilot operate by regurgitate portions of the training data, regardless of how small such regurgitation is, such doubt does not exist. The pirate bay case is a pretty clear example where regardless if people split files into tiny pieces, if the copying of file is just theoretical and not proven, even if there is no individual even being accused of doing the actually copying, courts can still find people guilty if they find the alternative theory of no copying at all unbelievable.
- michaelbrave 4y ago
  I'm not entirely sure that training a model on something counts as copying, it more or less recognizes and regurgitates patterns, something like x has such percent of chance to follow y such percent of chance of times when z keyword is used. Then uses that to recreate junk from more or less randomly generated static and finding patterns related to keyword. So it's not actually copied, it's closer to creating a chart or a graph representing the number of times a word was used.
  - belorn 4y ago
    If you type in something like "username = " and ask copilot to autocomplete it will follow its weights to find a username that in context has the highest value and then copy it. Unless a microsoft developer has manually gone into the software and written code to call a name generator for usernames and password generator for passwords, copilot will copy paste what it has in the training data. This was one of the many problems at launch since people did this for databases accounts, online accounts and so on. Copilot like all its similar created software can not create novel concepts. It can only copy and rearrange.
puffoflogic 4y ago
As a putative member of the plaintiff class, I feel my interests are far better aligned with GitHub than the representative plaintiffs. Unfortunately I am aware of no way to take such a position before the court to prevent class certification, other than perhaps a very heavyweight process like intervening. If I merely opted out of a settlement, the damage would already have been done: in effect, the plaintiffs have hijacked my identity (and more importantly, thousands of others) for their own personal benefit. Also, as always, the case is frivolous because the plaintiff class has each and every one (including myself) either licensed their code to GitHub under a license which does not require attribution (under GutHub EULA) or are the victims of a third party purporting to license the works to GitHub, without rights to do so.
- KRAKRISMOTT 4y ago
  You can send the court a letter.

3 points by sam345 4y ago

> The plaintiffs allege that Codex and Copilot violates the open-source programmers’ copyrights because Copilot will sometimes regurgitate code it was trained on This misunderstanding seems to be in the core of the discussion. One side thinking that copilot can write novel and unique new code using the mechanics of the programming language. The other side do not agree, instead seeing Copilot as simply an complex algorithm that regurgitate portions of the training data depending on context. The article do later go into a bit of discussion if those portions of training data can be considered copyrightable, but there are precedent. If you take a copyrighted work and split it into many small pieces, each much smaller than any copyrightable work, they can still be considered protected. Bittorent is one example, but there has also been examples where paintings has been cut into physical pieces and made into new creative work. In both people have argued that the individual portions aren't copyrightable and lost that argument in court. > Proving that all or even some of the code the purported class members wrote are copyrightable is not possible to do on a class-wide basis. It would require a judge or jury to opine on an instance-by-instance basis whether a given class members’ purportedly copyrighted code was original enough to qualify for a copyright. The pirate bay founders made a similar argument. Their lawyers argue that the court would had to first opine on an instance-by-instance basis for each work that was claimed to be pirated. The judges however said that such instance-by-instance work isn't necessary unless there was doubted that copyrighted work had been copied through the site. If Copilot operate by regurgitate portions of the training data, regardless of how small such regurgitation is, such doubt does not exist. The pirate bay case is a pretty clear example where regardless if people split files into tiny pieces, if the copying of file is just theoretical and not proven, even if there is no individual even being accused of doing the actually copying, courts can still find people guilty if they find the alternative theory of no copying at all unbelievable.

michaelbrave 4y ago
I'm not entirely sure that training a model on something counts as copying, it more or less recognizes and regurgitates patterns, something like x has such percent of chance to follow y such percent of chance of times when z keyword is used. Then uses that to recreate junk from more or less randomly generated static and finding patterns related to keyword. So it's not actually copied, it's closer to creating a chart or a graph representing the number of times a word was used.
- belorn 4y ago
  If you type in something like "username = " and ask copilot to autocomplete it will follow its weights to find a username that in context has the highest value and then copy it. Unless a microsoft developer has manually gone into the software and written code to call a name generator for usernames and password generator for passwords, copilot will copy paste what it has in the training data. This was one of the many problems at launch since people did this for databases accounts, online accounts and so on. Copilot like all its similar created software can not create novel concepts. It can only copy and rearrange.

puffoflogic 4y ago

As a putative member of the plaintiff class, I feel my interests are far better aligned with GitHub than the representative plaintiffs. Unfortunately I am aware of no way to take such a position before the court to prevent class certification, other than perhaps a very heavyweight process like intervening. If I merely opted out of a settlement, the damage would already have been done: in effect, the plaintiffs have hijacked my identity (and more importantly, thousands of others) for their own personal benefit. Also, as always, the case is frivolous because the plaintiff class has each and every one (including myself) either licensed their code to GitHub under a license which does not require attribution (under GutHub EULA) or are the victims of a third party purporting to license the works to GitHub, without rights to do so.

KRAKRISMOTT 4y ago
You can send the court a letter.