[email protected] (Richard Kenner) writes:
>> IMO there's legal questions we cannot really answer and there's
>> common sense. Taking
>>
>> - length_factor = size_int (vect_factor);
>> + length_factor = size_int (vect_factor - 1);
>>
>> how can you not literally do the very same upon suggestion from
>> LLM or a friend?
>
> Sure, but that change is not copyrightable whether written by a human
> or LLM because it's functional, not creative. So we needn't be concerned
> about scenarios like this.
>
>> My common legal-sense also says that changes like above are obviously
>> a derivative of GCC and not a derivative of sth else and thus
>> "legally insignificant".
>
> Even if the change is large enough to be legally significant, I think
> it's clear that it's a derived work from GCC.
>
>> But yes, I'm not really happy with the wording of our AI policy.
>
> Nor am I because of this issue among others. The policy allows for the
> possibility that the use of an LLM somehow "cancels" that derivation. My
> own view, and the view of many others, is that, although this particular
> issue hasn't been litigated, it's highly unlikely that it will be decided
> in that way. (If an LLM translates a copyrighted book, does that mean the
> result can be used freely? I doubt it).
I've deliberately not participated in the discussion about the policy
until now, and I don't want to offer a personal opinion. But since the
comments in this thread seem to have been against the policy, and since
I can see both sides, I'll play devil's advocate.
My understanding is that the policy is written the way it is because
of copyright concerns. Currently, for each author of a patch:
(a) that author's work needs to be covered by a copyright assignment
to the FSF or
(b) that author needs to certify that their contribution falls under
the DCO or
(c) that patch needs to be neglible for copyright purposes
Most people use LLMs as essentially black boxes, so it's hard to see
how anyone could say that the output of an LLM falls within (a) and (b).
Dominic emphasised that they'd reviewed Claude's output at every stage.
And that's important of course. But it's also the role played by patch
review. So in a sense, taking code from an LLM, reviewing it, then
submitting it, is similar to taking code from an unknown source and
having two reviewers review it.
So I can see the argument that taking code from an LLM and checking it
is similar to pulling code from an arbitrary github repo (with no
license or copyright info), adjusting it to GCC's interfaces,
reviewing it, and submitting that.
It seems like it would be difficult to relax the policy much without
also relaxing the copyright/origin rules for contributions that don't
use LLMs.
Like I say, just playing devil's advocate.
Richard
