Brandon,

By extension, yes Julian _is_ the one driving this policy: https://gitlab.haskell.org/ghc/ghc/-/issues/27433.

Julian,

You seem to want to misunderstand me. I have said nowhere that the following labeling is on its own hostile towards contributors.
Whether or not it is

We prefer contributions with non-trivial LLM assistance to be labeled LLM-assisted.
or
>
 We require contributions with non-trivial LLM assistance to be labeled LLM-assisted.

is pretty much irrelevant to me.  I've defaulted to prefer, simply because the ask is non-enforceable, and again, I feel very uncomfortable explicitly dictating to others what to do.  The implicit judgement from the policy is not.  You are fine with the policy because it reinforces your fundamental belief (I assume) that LLMs are bad, horrible, corrupting, and society damaging tech-bro inventions.  I am not, because it tells people that they are less welcome if they use LLMs in any form ("we strongly prefer"), I object especially to P3's wording in the document.  You do not seem to (want to?) understand that this can be understood as hostile language.  The policy ends up prescribing a process of how people should act, instead of describing what we want.

The policy exaggerated states: we want you to write code by hand without the use of assistive technology, and if you end up writing code with assistive technology, we think less of you. 

Now this (I assume) very much aligns with your worldview.  But I fundamentally disagree with that.  It's not inclusive.  

Just FTR so you can refer back to this:
(1) Easy to review, high quality MRs. 🤝
(2) Full responsibility. 🤝
(3) Declaration/Labeling 🤝 -- although, again I don't think LLMs are special here, it's a class of assistive technologies.
(4) Segregating contributors/judging them based on their preferences ❌

Best,
 Moritz


On Thu, 16 Jul 2026 at 13:13, Julian Ospald via ghc-devs <ghc-devs@haskell.org> wrote:
> *We prefer contributions with non-trivial LLM assistance to be labeled LLM-assisted.*

I'm also starting to think this discussion has run its course. Are we now arguing about whether to say "prefer" or "must" and that difference is going to make the world whether people will still be contributing to GHC?

We can't force people to disclose anything. As you've explained in your earlier emails, we have to assume good faith. And now you're turning around and say "oh, but we can't assume good faith, so it's not enforcable anyway". You gotta pick one.

I think we should:

- require LLM contributions to be labeled as such: this is important for context and is part of respectful collaboration
- assume good faith: if people (actively) mislabel their contributions, we will notice and ask them to correct that

I think a project is well within its right to make this demand and I do absolutely disagree with you that this labelling act in its own is hostile towards contributors.
_______________________________________________
ghc-devs mailing list -- ghc-devs@haskell.org
To unsubscribe send an email to ghc-devs-leave@haskell.org