Quote:
Originally Posted by multiweb
|
Hi Marc,
AI assisted code writing is absolutely going to become a hundred billion dollar market in the years ahead.
However, GitHub's Copilot is already controversial when it was commercialized.
Last year a copyright lawsuit was filed against them.
Claim is that Copilot was trained on some code that was copyrighted under
various open source licenses. So for example if one those licences had
been the restrictive greater GPL, the question in law becomes is the
outputted code from Copilot tainted with GPL? If you then use that
code as part of your company's closed source software, have you
potentially tainted that as well?
Stories and opinion pieces here :-
https://www.theregister.com/2022/11/...pilot_opinion/
https://felixreda.eu/2021/07/github-...our-copyright/
https://www.theverge.com/2022/11/8/2...-training-data
Now where it became really controversial is in December 2022 when
Tim Davis, a computer science professor at Texas A&M University, found
he could get Copilot to emit large chunks of his copyrighted code that it had
slurped up in the training set, but missing all of the attribution and
license comments.
https://techcrunch.com/2022/12/08/gi...in-unresolved/
Quote:
|
Originally Posted by Tim Davis Twiiter post
@github copilot, with "public code" blocked, emits large chunks of my copyrighted code, with no attribution, no LGPL license. For example, the simple prompt "sparse matrix transpose, cs_" produces my cs_transpose in CSparse. My code on left, github on right. Not OK.
|
Screenshot image below.
Now any professional software engineer called upon to give expert
testimony is going to put their hand on a copy of K&R and swear to tell the
truth and state that in their professional opinion, that's extremely likely to be
a lift of Davis's code with trivial changes.