Hi Carl,
Am 01.06.26 um 02:29 schrieb Carl Sorensen:
I believe that most new contributors start with relatively low-line-count
patches. And only get.to high-line-count patches when they have
understanding and capability with the lilypond codename.
AI-generated patches can easily ne larger, but there is no.guarantwe that
the AI agents actually reflect appropriate understanding of the code base.
Well, it's probably no coincidence that the two instances of proposed
AI-assisted MRs that I am aware of both shared the features of
- the authors having contributed no or very few changes before,
- large amounts of added lines,
- tackling the most famous long standing issue (#34).
Issue #34 has obviously never been resolved even though it has been
discussed again and again over the years - and, if I may say so,
discussed by people with (in my impression) immense skill and ability.
The optimistic view of this would be that the new technology finally
enables us to solve problems that have baffled human problem-solving
abilities for 20 years.
The pessimistic view would be that the ease of generating seemingly
working code that AI technology provides inspires overconfidence.
A more realistic optimistic stance might be the following: AI technology
enables a far greater number of people than before to try their skills
at hard problems than before since it helps deal with the significant
hurdle of understanding enough of the very complicated LilyPond codebase
to get started in the first place.
Lukas
