That said, lets say there's a new model that explicitly excluded closed source and copyleft licenses. Well, the MIT, MPL, Apache, BSD- they all say you can't strip their licensing off.
Okay, so to get to the spirit of your question, lets say Github managed to program a model that worked using only their own code or code that was explicitly put in the public domain. If Github managed to reproduce code that wasn't in the training set, then it can't be accused of copying it. At that point the argument could be made that it independently created it.
At the same time algorithms can't be copyrighted, but implementations of an algorithm can be, so if Github was basically just spitting out an algorithm that just happened to be implemented similarly to how some other code it wasn't trained on implemented it, then I would say there was no copyright violation.
If the comment is something like
//check fromIndex is greater than toIndex
then that is not any more individualistic or different than the actual function. Sadly, many people comment like this, on the other hand if it reproduced a comment with typos or something more complicated like
/* this hack is because Firefox's implementation of SVG z-indexing does not match how Chrome or Safari does it - please read this article ...url...*/
then yeah, then you would have something
https://twitter.com/StefanKarpinski/status/14109710611816816...
So you literally can't make it produce functionally identical but not verbatim identical code. It doesn't understand that the two are equivalent.
Also, such "functionally identical but not violating copyright" transformation is not possible to do, both given the complexity of the problem and the sheer volume of the data.
And training it on some simplistically obfuscated code wouldn't help - all it would learn would be production of obfuscated code. Not useful for the intended use.
it doesn't need to understand the way a human might do the understanding.
The pattern that the LLM managed to extract could include the structure, rather than the pure text. And in reproducing the structure, the LLM can replace the variable names but keep the structure intact.
I am not sure if copilot is able to do this, but chatGPT was somewhat able to (if imperfectly at the moment).
But it does - similar but not identical code are closer in the embedding space
That's what patents are for.
