Hi Mark,
On Tue, Feb 10, 2026 at 8:25 AM Mark Wielaard <[email protected]> wrote:
>
> Hi Aaron,
>
> On Sun, 2026-02-08 at 19:43 -0500, Aaron Merey wrote:
> > In December we discussed whether elfutils should permit LLM-generated
> > contributions [1]. The conclusion of this discussion was that elfutils
> > should not permit this. Reasons for this decision include uncertainty
> > around the copyright status of LLM-generated content. Below is a
> > revised proposal based on these discussions.
> >
> > If accepted we can include this policy in the CONTRIBUTING document in
> > the elfutils source directory. The text of this policy is modelled
> > after the Binutils [2] and QEMU [3] LLM policies.
>
> Thanks for drafting this. It looks good to me. One small suggestion
> wording "tweak" below.
>
> > Policy on the Use of LLM-generated Content
> >
> > The elfutils project does not currently accept contributions
> > containing output generated by Large Language Models (LLMs) [4]. Use
> > of LLMs to research, analyze or debug a contribution is allowed as
> > long as no LLM-generated output is included in the contribution.
> >
> > There are two exceptions where LLM output may be included in a contribution:
> >
> > (1) The output consists solely of trivial changes such as spelling or
> > code formatting.
> >
> > (2) The LLM assists in writing a contribution but does not author any
> > of the content. This includes accessibility-related uses of LLMs
> > involving speech-to-text, for example.
>
> I understand what you are saying here, but might not use the words
> "writing" and "author" (which feel a little as if we are "personifying"
> the the llm). How about using the words "researching" and "generating":
>
> (2) The LLM assists in the research of an contribution but does
> not generate any of the content. This includes accessibility-related
> uses of LLMs involving speech-to-text, for example.
I think this wording is an improvement but what do you think of
mentioning "transcription" as well?
(2) The LLM assists in the research or transcription of a contribution but
does not generate any of the content. This includes accessibility-related
uses of LLMs involving speech-to-text, for example.
Speech-to-text transcription is one of the accessibility use cases I
had in mind when I tried to distinguish between authorship and
writing.
Aaron
>
> (Note English isn't my first language, this might not be gramatically
> correct.)
>
> > Contributors are not required to disclose the use of LLMs for these
> > purposes.
> >
> > This policy may be reviewed or updated when the copyright status and
> > Developer's Certificate of Origin (DCO) compatibility of LLM-generated
> > output is clarified.
> >
> > [1]
> > https://inbox.sourceware.org/elfutils-devel/cajdtp-r9m7uvfggoq20_4k8ooja4lvmuz3x8tzrhq-+r5aa...@mail.gmail.com/T/#m52aef56465c8bbe6d4fe0fda6487add9efb4f857
> > [2] https://sourceware.org/binutils/wiki/LLM_Generated_Content
> > [3]
> > https://www.qemu.org/docs/master/devel/code-provenance.html#use-of-ai-generated-content
> > [4] https://en.wikipedia.org/wiki/Large_language_model
>
> I would be OK with you adding this to CONTRIBUTING with or without the
> wordsmithing above.
>
> Cheers,
>
> Mark
>
