Julian,

I've quoted your email on purpose.  You clearly state that LLM usage for
you is a signal of distrust.  The policy itself has a strong bias towards
human written code.  It does not protect people who use LLMs in any form;
it tolerates LLM usage, and it classifies LLM users.  It protects, and
codifies a bias, while not outright banning LLMs.  I have no issue with
attribution, you'll find plenty of comments/notes with attributions from me.

Again, my point is that it sets incentives that I believe are not aligned
with your preferred outcome.

Best,
 Moritz

On Thu, 16 Jul 2026 at 11:34, Julian Ospald via ghc-devs <
[email protected]> wrote:

> > Truthfully discloses LLM generation => Additional scrutiny or stricter
> review, because we distrust you
>
> This is not what the policy says.
>
> Even pre-LLM it was common sense and common courtesy to disclose when you
> copy-pasted code from other sources, be it stackoverflow or some other
> project. I do that and it's also a legal requirement for a fair number of
> open source licenses and it doesn't matter whether you reviewed all of
> that... it doesn't absolve you from declaring it. It also helps the
> reviewer get more context: where do the ideas originate from, how much
> effort was potentially involved in certain parts of the patches... those
> are all fuzzy intuitions... not final judgements.
>
> Whether LLM generated (or copy-pasted) code needs more scrutiny or not
> depends on a lot of factors. It may very well require more scrutiny, it
> depends on the scope, the workflow, the patch owner, etc. Withholding that
> information is simply poor collaboration. You're interacting with another
> human and their time and energy. Your reviewer may not care or they may. We
> won't be able to impose opinions on anyone, but this disclosure is
> essential for other people to get the context they need to conduct a
> review. It doesn't say whether the reviewer will be negatively biased
> against AI or not. This discussion won't change those positions. But it
> will set boundaries as to how they can treat you: they won't be able to
> reject your patch just because it's AI generated.
>
> So in a way, this policy is trying to protect both sides.
> _______________________________________________
> ghc-devs mailing list -- [email protected]
> To unsubscribe send an email to [email protected]
>
_______________________________________________
ghc-devs mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to