José E. Marchesi wrote:
> I would personally reject such a patch unless it is entirely legally
> insignificant.

Yes, that's essentially what we do here, by the Gnulib LLM policy
https://gitweb.git.savannah.gnu.org/gitweb/?p=gnulib.git;a=blob;f=HACKING;h=72c9895c0f407a28f5c5c82b0173ccf6e89b4429;hb=HEAD#l226
 .

> I would avoid using Co-authored-by this way, and use Assisted-by (or
> similar) instead.

I agree. And still, even
  Assisted-by: Claude
or
  Helped-by: Claude
are ambiguous, because it does not tell what amount of code came from the LLM.

To help current and future evaluations of such a patch, it would be useful
if the submitter states — either in the git commit message or in the mailing
list post — explicitly:
  "I used LLM 'Claude'. The output of the LLM was textual only, no code.
   (I ensured this by requesting no code, in my prompt.)
   I produced the code contained in this patch myself."

Bruno




Reply via email to