Re: [MAINTAINERS SUMMIT] Annotating patches containing AI-assisted code

[Laurent Pinchart <laurent.pinchart@ideasonboard.com>]

On Mon, Aug 18, 2025 at 09:47:04PM +0200, Jiri Kosina wrote:
> On Mon, 18 Aug 2025, Rafael J. Wysocki wrote:
> 
> > I tend to agree that such annotations might be useful as heads-up
> > markers for maintainers if nothing else, but what about missing
> > annotations?
> > 
> > Is there a generally feasible way to figure out that they are missing?
> 
> Maybe we can use some LLM to help us decide whether the code has been 
> written by a human or LLM :P
> 
> >  And if that can be done, "suspicious" changes may as well be caught
> > this way, so why would the annotations be required after all?
> 
> I am not sure whether we have more options than documenting this 
> requirement, and then work with our usual tool, which is building trust 
> (or lack of thereof) in the individual submitters.

At this point, I would expect contributors to not mention that code has
been generated by an LLM for three reasons, in decreasing order of
frequency:

- Because we don't tell them they need to (that's the current situation)
- Becuase they don't know they need to (people don't read documentation,
  so we'll have to find ways to get the message through)
- Because they maliciously decide to breach the rule

I'm not too concerned about the third reason for the time being, as we
have way more developers acting in good faith than bad faith. If that
changes, we'll have to figure out how to handle the problem.

-- 
Regards,

Laurent Pinchart