On Sun, Jul 26, 2026 at 11:23:29PM +0200, amindfv--- via ghc-devs wrote:
> I'll tell you that my interaction on this list with discussion that
> was LLM-generated (even with full responsibility taken,
> hand-selection, etc.) has affected my already dim view negatively.
>
> In principle we can say "you must stand behind every line." Standing
> behind every line does not mean that the whole process has not been
> made worse and unwelcoming. The ability to generate HTML tables
> summarizing one's point without considering the other fellow's point
> is not an unvarnished good.

I understand you are talking about Moritz's recent emails to this
thread.  I have to say, my experience reading them has made me more
sympathetic to those who are concerned that, with LLMs, contributions
to GHC will go up in volume and down in average quality.  I have not
really been able to understand the point that Moritz is making, have
not read many of his recent emails in full, and although I understand
he's more or less seeing things from a similar point of view to me I
don't think he's successfully persuading anyone.

The argument I would like to see is one about how LLMs can increase
code quality, decrease workload for an equivalent result, and make
everyone's experience using GHC more enjoyable, not less (modulo
discomfort caused by moral concerns about use of LLMs *at all*,
tackling which I doubt would be a productive use of my time).

> I think your argument -- well, sometimes it's as good or better --
> is eminently reasonable.

To be clear, although that is something I believe (in fact I think
given its use of the word "sometimes" it's a very weak claim) but it
is not an argument I have made yet.

> Yet in aggregate, expressing only a
> "recommendation" for hand-written code/docs/discussion will have
> very obvious negative effects. It already has. This is precisely why
> I advocate the policy being preference-based instead of a ban or
> full permissiveness.

> I think for many people it's both, but speaking for myself primarily the 
> former.

That's good to know.  My hunch, given what I've seen of Jaro writing
on the matter in various fora, is that the latter is also a
significant source of discomfort for him (possibly to a greater degree
than the former) and further I guess that the same applies to some
others in our community.  This distinction is important because the
former objection is, in principle, resolvable by an appropriate choice
of tooling, procedures and process but the latter is not!  In fact,
there might be ways of ameliorating the former source of discomfort by
*increasing* LLM use, but that certainly couldn't ameliorate the
latter source.

Tom
_______________________________________________
ghc-devs mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to