All public GitHub code was used in training Copilot

Re:All public GitHub code was used in training Copilot by Diggsey on Thursday July 08, 2021 @11:05AM (#27772881)

Also NAL, but I think there's far more of a case that users of Copilot might violate copyright rather than Copilot itself:

- Only a very small proportion of Copilot generated code is reproduced verbatim, so if you specifically built a product just from copied-verbatim code, your act of selecting and combining those pieces of copyrighted code would be creating a derivative work.

- GitHub is not selling the copyrighted code, they are selling the tool itself. Google is literally the same thing: you could theoretically create a product by googling for prefixes of copyrighted code and then copying the remainder straight out of the search results. It's you who would be violating copyright, not google.

[ Parent | on HN ]

Re:All public GitHub code was used in training Copilot by ghoward on Thursday July 08, 2021 @11:14AM (#27773018)

I think there is an argument to be made that Copilot is producing derivative code, though. It may produce copies verbatim, and that's a violation, but far more often, it produces a mixture of things it was trained on, most of which probably have some sort of license requiring attribution at the very least.

[ Parent | on HN ]