[v2] Policy proposal for LLM-generated contributions

[Serhei Makarov]

On Wed, Feb 11, 2026, at 6:09 PM, Frank Ch. Eigler wrote:
> I fear that by being so nervous, we are going to exclude some people
> who could make useful contributions.  Productivity gains from
> AI-enhanced IDEs are widely reported - and I can personally testify to
> them.
>
> Also by being so nervous, we are going to motivate other people to
> just casually use LLMs in their work, but neither disclose this nor
> bother with even cursory license investigation, because it would just
> be punished by rejection.

Weighing in on this fairly late, but one thought seems worth pointing out.
(It took me a while to digest my own positions on this.)

I believe there are reasons specific to Elfutils for being more cautious/conservative,
outside of an evaluation of legal risk.

This project functions as a de-facto reference implementation for ELF and DWARF standard.
If I want to understand how DWARF works, this is a good place to look. Seeing working code gives greater confidence than reading standards documents that might be technically precise but confusing English.

Would referring to an elfutils implementation continue to be useful if the commit history and git blame showed an increasing proportion of LLM-generated stuff?

The LLM advocacy I'm able to find seems to neglect this kind of question, accepting some decline in quality with "well, the specs are now the canonical source of truth" and "well, the LLM can be asked to fix problems introduced by earlier use of the LLM".

-- 
All the best,
    Serhei