Hi Greg,

Greg Hogan <[email protected]> skribis:

> But the rule of thumb is losing nuance when summarizing the linked
> document. Oracle and GNU have the same goal, to own the copyright in
> order to sue (for profit and to protect the GPL, respectively). And
> the documentation is clear that simple contributions which separately
> are non-copyrightable can as a compilation become copyrightable. But
> Guix does not aggregate copyright, instead simply requiring licensing
> of the derivative work under the GPLv3. Since "legally significant"
> cannot have the same meaning within the Guix project, how are we to
> understand the use of this term in GCD 008? The text says "to ensure
> the contributor has a valid copyright claim on the code" but the rule
> of thumb seems to attempt to list non-copyrightable contributions.

Yes, I’m sorry: the wording was confusing (cannot have a copyright claim
on something that is not copyrightable).  Jason Conroy pointed out the
same problem and I pushed a fix.

>> I see two possibilities.
>>
>> 1. Clarify “encouragement”:
>>
>>   Examples of what would qualify as “encouraging the use of genAI”
>>   includes: providing project members with a commercial LLM
>>   subscription, communicating publicly on genAI-produced artifacts
>>   related to Guix (e.g., in blog posts), suggesting in pull requests and
>>   issues that people use genAI to solve their problem, and any similar
>>   action.
>>
>> 2. Drop “nor encourage use”.  I’m not a fan of this as it weakens the
>> rule, but maybe that’s okay in practice (hard for the project to
>> encourage without using).
>
> I would expect that references to non-free models would be disallowed,
> as with non-free software, but that generic discussion and help with
> permitted use would be allowed.

No; that’s not an example I gave above, and not one I have in mind.

One can refer to Claude, just like one can write about Blender, but it’s
just not the right forum for this.

> As elsewhere, applying the existing consensus project rules and
> guidelines to this new technology rather than creating a new
> regulation.

There are no existing rules on this topic though.

>> Right, exactly.  By “copying text” I meant pasting text: the human is in
>> the loop, but only to hit C-c and C-v (indistinguishable from an agent).
>> Would saying “pasting” instead of “copying” (I think that’s how Simon
>> phrased it) be clearer?
>>
>> What kind of “genAI actions” not covered by policy rule #3 do you have
>> in mind?
>
> Somewhat continuing from above, I expect the project would only run
> free models, and I can imagine a future where the user could request
> insight from that model on failed builds similar to guix-cuirass-bot.
> Or a QA-like system could evaluate logs for failed builds before
> retrying multiple times, conserving resources. The user would be
> taking action (such as following a link) to see the generative output.

I would rather not make a policy based on this hypothetical scenario
because AFAIK, “free models” (as in gratis) today perform poorly and
very slowly on commodity hardware.

> Copying vs pasting makes no difference to me, but I would prefer to
> have these examples and clarifications in the GCD text. Or the
> Originalists, Textualists, and Living Documentists must argue over the
> meaning.

Which examples and clarifications would you like to add?

>> Instead I would prefer plain text disclosure, a checkbox in the pull
>> request template, or a tag like ‘LLM-Assistance: debugging’.
>>
>> WDYT?
>
> Is a checkbox or tag sufficient for that use?

I think so: it’s enough for the reviewer to know that an LLM was used in
some way.

Thanks,
Ludo’.

Reply via email to