>>>>> "Gard" == Gard Spreemann <[email protected]> writes:
Gard> Again, I am not a lawyer, but this merely states that
Gard> Anthropic does not retain claim ownership over the output of
Gard> the model, not that you can necessarily claim copyright over
Gard> it.
I think there are two aspects to this:
1) Things no one can claim copyright to.
At least in the US, it may be that AI output cannot be copyrighted.
I don't know if the question of whether the output can be a derivative
of the prompt that produced it has been settled.
It may be that the prompt is expressive enough that you can claim
copyright because of the prompt.
But let's assume that novel Ai output (output that is not a derivative
of some other work used in training) cannot be copyrighted.
So what? I don't think that should block Debian. If it turns out that
some court later decides the work is public domain because it has no
copyrightable components, 1) public domain is DFSG free, and 2) public
domain is GPL compatible.
There may (likely are) be jurisdictions where the GPL can apply.
So in my mind the case of uncopyrightable output is not something we
need to consider because at worst we're claiming to our users a more
restrictive license than actually applies.
2) There's the case where some third party can claim copyright because
for example the output is a derivative of training data in which the
third party holds copyright.
In my mind this is a case where Debian needs to make a determination
about acceptable risk just like we do with things like patents.
We aren't going to settle all the questions here; it will take years for
things to work through the courts, and different parts of the world may
decide differently.
This is not the first case where Debian has had to make a risk
evaluation--not even the first case related to copyright.
It seems like accepting no risk is probably the wrong answer,
particularly in light of the recent GR results.
But that doesn't mean we should accept all risk.
signature.asc
Description: PGP signature
