Am Di 21.07.2026 09:15 schrieb Simon Peyton Jones via ghc-devs:
There has been a lot of discussion about GHC’s LLM policy on this GHC thread:
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-SH8/edit?usp=sharing>
[…] Any final thoughts?
I’ve read the current version of the draft in its entirety. My assessment is still more or less the same: I don’t agree with multiple things in this policy, but it seems to me that using such a policy for the time being is the only way to somehow move forward together. I sense that the text has notably moved towards the “pro-LLM” side. Sometimes this led to outright wrong statements, like the following:
If you carefully review and understand every line of code or documentation produced with LLM assistance, revise it where necessary, and take full responsibility for every line, that constitutes human authorship.
No, it doesn’t. An author is a person who *created* the work. If the work was assembled by an LLM, based on only guidelines by the person in question, then said person acted only as a reviewer, not as an author. Maybe the more LLM-friendly tone will enable some of the pro-LLM people to at least tolerate this policy. I think that we shouldn’t move even further into the pro-LLM direction to please even more people. All the best, Wolfgang