Re: Merging AI-generated program code

Am 01.06.26 um 11:07 schrieb Lukas-Fabian Moser:
>
> Hi Arno,
>
> Am 01.06.26 um 10:36 schrieb Arno Waschk via Discussions on LilyPond 
> development:
>> This seems wrong at least at two points:
>>
>> 1. The automatic assumption that any support of AI makes a 
>> contributor not read his contribution, not understand his 
>> contribution, and the whole thing ununderstandable to a degree which 
>> can not be remedied in the same way as hand written code is 
>> completely irrational. AI code can be bad. Manual code can be bad. I 
>> produced both kinds. Reasonable people read their code before 
>> submitting, or, say, try to put it in shape. Assuming people lose 
>> their reasonable minds by using an AI  summary of a local MR index 
>> is, sorry, ridiculous.
>
> Nowhere did I state that.
>
> It's a fact that AI assistance makes it /possible/ to generate large 
> amounts of working code that the human contributor doesn't fully 
> understand. This was practically impossible before, so it creates a 
> new situation.
>
I can easily produce code manually, and not understand it. Or worse, 
live in the illusion to understand what i am doing, but being totally 
wrong in this.

> This becomes a problem since a system (like with LilyPond) that relies 
> on human code review (together with a very small pool of active 
> reviewers) can easily be overwhelmed by the amount of code that can be 
> produced with the new technology. So we must find reasonable 
> guidelines how to deal with the new technology.
>
Currently the amount of MRs, especially when not counting those which 
seem to come from what looks to me like being the core lilypond team, 
which rightfully could be trusted more than MRs from outsiders, 
obviously, is rather underwhelming.

(I know that this is different in other projects. One could learn from 
those *before* assuming problems which either do not exist, or can be 
dealt with in case they might exists at any given point)

> I view Dan's question as an honest attempt do find a solution. I'm 
> surprised that you construed his e-mail (which also contained the 
> sentence "I just wouldn't want the uncertainty to last so long that a 
> capable contributor gets frustrated and leaves.") as an expression of 
> "academic AI hate" or "something personal".
>
I cannot separate this email from other communication which did already 
happen. If "capable" was for me, nice, but then all the other explicit 
and implicit pejoratives against persons and code apply to me too. And 
enough details were mentioned which can only apply to me, or rather my 
MR !3058
>
> Furthermore, I also explicitly pointed out that there are various 
> kinds of AI assistance; so my opinion was considerably more nuanced 
> than you represent it here.
>
>> 2. What you wrote basically means, that code can be trusted blindly 
>> once it was written by hand. This is but a joke.
>
> No, it doesn't; please don't misrepresent what I wrote.
>
> On the contrary, what I wrote was this:
>
> "This is not just a question about AI contributions: It means that 
> LilyPond also shouldn't contain human-written code of the "I added 
> that line and then the problem somehow went away, knock on wood" type. 
> It's of course hard to enforce this, but a thorough review where 
> questions can be raised and must be dealt with makes it more probable."
>
> Lukas
>
If this is to mean that any code needs to be read by contributors and 
reviewers, we can agree on this. But then any discussion about how much 
bias is justified against code which comes with whatever support from 
whatever kind of AI (which is how i understand the wordings here) is 
completely obsolete.

And i hope that on can understand that it adds only to the frustration 
if one spends quite some time on 20 years of silently cursing about #34 
but never complaining (always thinking "i had better write code than 
complain"), observing that other attempts to solve this by even the most 
lilypond experienced coders did not materialize over many years, 
external attempts also gave up (e. g. !3002) and one finally sees a way 
to propose a solution (nota bene: not claiming a 100% one) and hands it 
in, expressedly as an unfinished proposal of an idea, with some working 
(!) code but with the question if before all any further investment into 
this makes sense in the eyes of the core developpers and project owners, 
is only met by "the git history is not perfectly clean, we won't 
evaluate even the concept" or "there was some assistance by AI, so we 
are to assume that he never read his code and doesn't understand it 
anyway" and ptraonising personal emails about how much rude tone of 
developers or reviewers one should be "conciliant" against, and 
pseudo-academic meta discussions like this one, or about if creating 
code is allowed to change from how it was the ideology in the 80s.


Maybe 10% of the communication was in one or the other way constructive. 
Thanks to Han-Wen in this context.


Maybe this is lilypond's strategy of keeping the amount of external MRs low.

Might work.

Arno