How to deal with fully LLM-generated merge requests

[Niccolò Ve]

Hi,

I like the policy Nate proposed and overall it's a +1 from me.

Personally I would also like to know which models have been used on a
certain patch, though; it's not a matter of things messing up in its own
way, but rather models having different strengths, even if the
contributor effectively balanced for that, it's nonetheless good to know
what to look for especially when reviewing. Not a dealbreaker for me though.

Nicco

Il giorno gio 10 set 2026 alle ore 21:17 Nate Graham <nate at kde.org> ha
scritto:

> On 9/10/26 1:02 PM, Lasath Fernando wrote:
> >
> > On Thu, Sep 10, 2026, at 13:33, Nate Graham wrote:
> >
> >> 5. Don't add "Assisted-by: [some LLM]" tags to your commits; it's just
> >> free advertising for the LLM's provider.
> >>
> >
> > I humbly request the opposite of this. Doesn't have to be in the commit
> as a co-author, but we should require people to mention the exact model
> they used at the bottom of the MR description (e.g. Sonnet 4.5 or GPT
> 5.6-Sol). I spend 90% of my day-job reviewing AI-assisted code, and it's
> extremely useful to know *which* LLM was used for two main reasons:
> >      1. It's much easier to determine how much thought the human has put
> in if you're familiar with the model they used
> >      2. Different models have very different things I watch for when
> reviewing
> >
> > I don't know if many other KDE contributors' employers are as Gung-Ho
> about AI use as mine is, but it'll be useful for the people who are used to
> reviewing AI code, and (I hope) doesn't hurt anyone else.
>
> If the exact model used is a relevant data point, then the contributor
> would have already violated the "don't vibe-code" guideline, no?
>
> As the guidelines say, if you're using an LLM responsibly, nobody should
> be able to tell you did. If it's so obvious that someone wants to know
> what model you used because they know exactly how each model messes
> things up in its own way, you did a bad job.
>
>
> Nate
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/plasma-devel/attachments/20260910/383ffaa6/attachment.htm>