Regretfully, and with full recognition of the need to balance the preferences of many developers, I think the removal of a statement of preference for human-written code takes the heart out of this document. It would cause me to withdraw my support, for whatever that's worth. My perspective is that of a very infrequent contributor who follows development on the Gitlab pretty closely, and values being able to do so. I understand the goal is to be as neutral as possible, but let me try and make the case that even putting aside the social and environmental (etc.) concerns, there is not an easy neutral position waiting for us: Take a very easy-to-imagine scenario: a person opens an MR, which begins with: "Here's some code. It's been generated using AI, as has its documentation. I stand by it." The revised draft policy seems to assume that that only people whose business that is are the author and whoever's willing to review the proposed changes. I feel it would be a great loss to arrive at the GHC Gitlab to read about a change of importance to me, and have to wade through a long document alternating between "This code has been written with LLMs. I stand by it." and "This review has been written with LLMs. I stand by it." And in the end, documentation will be something many will read, even if they have a strong distaste for AI writing style. A developer taking responsibility for every word may still contribute to a reduction in quality of text that many of us must later read. Already in discussing this policy, I've had to read far more LLM-generated text than I'd have liked. I had a distinct sensation that I was being asked to re-read the same points as had been previously made, padded out in length by an LLM. We then got confirmation of that at the end of a 792-word email (that's 4 double-spaced pages). I'd find it ironic if the LLM usage policy were changed substantially because of a large number of emails, from a small number of people, with a large contribution of "assistive technology." Everyones' points shold be considered and valued, but there's a risk that the people in favor of text-generating machines may drown out the other voices by word count. If the preference of GHC developers is to use LLMs, then this policy may be fine. However, I thought the previous policy -- which was very open to LLM contributions but made a statement of preference for a community of human developers speaking to each other through code and prose -- was a much better basis for community standards in a time where LLMs make it very difficult to tell good from "good enough," and another person from "person enough." - Tom
On 07/21/2026 10:15 AM CEST Simon Peyton Jones via ghc-devs <ghc-devs@haskell.org> wrote:
Friends
There has been a lot of discussion about GHC's LLM policy on this GHC thread (https://mailman.haskell.org/archives/list/ghc-devs@haskell.org/thread/ZTHFWO...).
I have made quite few revisions in the light of this and other conversations. Here is the draft again (https://docs.google.com/document/d/1XxP8NXnbgr6faxXBXKw75zgoxIg-SzGv-Cvixyi-...), I hope in a close to final form. Any final thoughts?
thanks
Simon _______________________________________________ ghc-devs mailing list -- ghc-devs@haskell.org To unsubscribe send an email to ghc-devs-leave@haskell.org