Re: Merging AI-generated program code

Am 31.05.26 um 22:23 schrieb Dan Eble:

Hello Dan,

probably this refers also to me so i guess i need to throw in some things.

> There was recently a thread on AI-generated documentation:
>
> https://lists.gnu.org/archive/html/lilypond-devel/2026-05/msg00025.html
>
> ---
>
> Now, I want to ask whether anyone sees any issue with merging 
> AI-generated program code.  This is somewhat urgent in the sense that 
> we have such an MR in draft, though of course we could delay it until 
> there is certainty.  I just wouldn't want the uncertainty to last so 
> long that a capable contributor gets frustrated and leaves.
As least for the one MR draft from my side there is enough reasons for 
frustration already, but the project taking time to read and evaluate 
the proposed code would never be a reason for frustration. Except maybe 
if it comes to decades ;-)
>
> To my knowledge (and I have been paying only minimal attention), the 
> FSF views AI-assisted contributions to GNU projects as potentially 
> problematic but has not established a policy.
If ideology guidance is needed if would chime in with the well known 
opinion of Linus Torvalds to that matter.
>
> As a reviewer, I strongly desire two things:
>
> 1. openness about the origin of the code I'm reviewing
> 2. accountability of the human submitter (not reviewers)
>    for the code that is merged
>
> For the MR that is in draft now, there were tells in the patch, but I 
> had to ask the submitter twice before he confirmed that it was 
> "AI-assisted."  To streamline this in the future, I propose 
> configuring a template for default MR descriptions something like this:
>
"I had to ask twice" is an interesting point here, when the inital 
proposal came with a clear question (which you confirmed to have 
understood, which i had to ask many many times and never got an answer, 
maybe apart from an implicit answer which i could project into the lines 
of this email.


> ##### Description
>
>     <!-- Describe your motivation and your work briefly
>     to orient reviewers.  If you have not described
>     your commits well, go back and do that first. -->
>
>     ##### Question
>
>     What percentage of this work is AI-generated?  <!-- 0-100 -->
>
> Do you think that would effectively address that specific concern?
>
No. It gives zero insight into point 2 from above, which i can see as 
the only valid point here. Where it has ZERO relevance how code was put 
together, no matter if it is bad manual work, clueless copying from 
stackoverflow, throwing dices or clueless vibe coding. Or working code wtf.


> ---
>
> Another situation arises at the confluence of these things:
>
> * Anyone with middle-class money and some patience can
>   rather easily submit a high-line-count patch that solves
>   a narrowly defined problem and passes regression tests.
>
If you think a problem was solved in an inefficient or over complicated 
way this should be evaluated. Where it has ZERO relevance how code was 
put together, no matter if it is bad manual work, clueless copying from 
stackoverflow, throwing dices or clueless vibe coding. Or working code wtf.
> * Unless the submitter first examines the generated code
>   carefully to confirm that it aligns with his intent more
>   broadly, that kind of generated code demands even more
>   of reviewers' time and attention than before.
>
Assuming that a submitter who said his code was "AI assisted" (where 
most of the assistance was in getting some kind of an overview over info 
scattered around in 3k MRs and 7k issues) did blindly generate anything 
and did not even care reading it, or did not care at all about if it 
"aligns with his intent"  is a blunt insult. No matter if this addresses 
me or some future submitters. Where it has ZERO relevance how code was 
put together, no matter if it is bad manual work, clueless copying from 
stackoverflow, throwing dices or clueless vibe coding. Or working code wtf.

I need to assume you mean me, as after you got my statement, that there 
was "AI-assistance" (not even talking about generation per se) you never 
asked a single line in terms of your valid point 2 form above.

So what's the point then?

Stale academic AI hate?

Or something personal?


Frustration maximized.

Arno

> * The number of active reviewers (those who have been
>   leaving comments) is currently small.
>
> * Our current "countdown" policy effectively treats
>   silence as approval, and moves changes forward as long
>   as there are no objections within a certain time.
>
> Therefore, I suggest adopting a new policy: AI-generated program code 
> does not automatically move forward without a human reviewer's 
> acknowledgment.  It should be full acknowledgment, not, for example, 
> "C++ LGTM; don't know about Scheme."
>
> It would fall to the "patch meister" to help people follow this policy 
> and to allow sensible exceptions, such as if a contributor with a good 
> record vouches for the quality of his own AI-generated submission in 
> an area where he has developed expertise.
>
> ---
>
> Is there any other guidance that we active contributors should follow 
> when dealing with AI-generated submissions?  The floor is open.
>
> Thanks,