things in alpine that have driven some doubt for me for some time:

- the overall governance situation (not fixed, but maybe marginally improving)

- apkv3 file format (i think it is good to use tarstreams, they are not dangerous)

- APKBUILDs just feel obsolete compared to melange

the fact that at least one developer stated that disclosing somebody else's publicly-documented conflict of interest is "doxxing"
i think my conclusion is that i want to build something better instead of continue to be a squeaky wheel that people find annoying

the fact that developers spend 90% of their time toiling with manually bumping packages, and this is why we "need" AI contributions

this stinks. it's bad engineering.

but most importantly, i want to build a community where *nobody* is infallible

one thing i have noticed, again and again and again, is that whenever anyone criticizes ncopa, people will blindly defend him

this is not good

and to be clear, i have not made a cogent argument for or against LLM contributions in the long tail. i think that we need to move slowly and focus on the fundamentals.

when the fundamentals are right, then the LLM argument looks a lot different -- time pressure which encourages LLM usage is reduced, leading to a more reasonable outcome

but what i do have conviction on is that i can't trust projects which have become hooked on commercial LLM coding.

the power imbalance is too great of a risk

@ariadne

This has been one of my biggest sources of frustration for the last few years... "AI" has sucked up all the oxygen in the room, and there's no time or energy left to do real engineering

@ariadne i’m ver unaware of the situation but what i understand from your post is the llm would be used to fix the consequence and not the root problem
@hypha you understand my concern correctly
@hypha @ariadne That's basically always the case when someone thinks LLMs will solve their problem.

@dalias @hypha @ariadne to be fair, humans do have a tendency to focus on "new shiny" and assume it will solve all their problems. LLMs are just the latest focus for some.

I guess what is more disturbing with LLMs (as already highlighted): they are owned by corporations who are objectively evil (not specific to AI companies, just a general comment on overly large companies in general).

@ariadne Honestly that sounds like an imminently automatable thing, not something you need an AI for though
@krutonium it *is* automatable. i demonstrated how at the previous company i worked for.
@ariadne Then why is ANYBODY spending tokens on that lmao.

@ariadne 90% of the work should be reviewing the changes to packages before bumping them to get at least some degree of confidence that there is nothing unsafe/breaking/hostile in the new version, and that's something that can't be automated.

Whether or not they automate the mechanical act of bumping, that act should be a tiny portion of the actual work of maintaining a distro package.

@dalias distros generally don't do that level of review, we assume good faith from upstreams unless demonstrated that we shouldn't.

@ariadne @dalias

You can assume good faith and still need to check if a downstream service/config/patch file needs to change. Stuff like that is not always in release notes (if they exist at all).

@ariadne fucking mood, right there
@ariadne i love your verve