Post by @zzzeek@hachyderm.io

@mitsuhiko @glyph I've unfortunately never had anyone who was sending me low quality code contributions ever go on to become a productive member of my projects . My successful contributors all started out being careful and rigorous from the start. Then* they go off to do other things, which is good for them (I of course get to maintain their [pretty good] code for all perpetuity unless it gets rewritten for whatever reason). This is probably very unfortunate on my part but I don't feel much interest in acting as a brief stepping stone for good programmers and a babysitter for bad ones. I just need code written

*I have a single co-maintainer who has stuck around and he is spectacular. We interact around the clock

10 Replies

0 Boosts

Armin Ronacher

@mitsuhiko@hachyderm.io 3 hr. ago

@zzzeek @glyph This might very well be the case, but you need to recognize that we're in a period of learning. We have these super powerful machines, but we don't yet know how to use them productively. But the solution can't be to opt out, we need to figure out how to use them.

An Open Source project that opts out of this change, is just going to defer that learning process.

I have seen this over and over now. The core of the issue is that if a project's leadership takes a hard line against it, that goes as far as not dabbling with the technology at all, then it becomes really hard because of the asymmetry that this creates.

Hynek Schlawack

@hynek@mastodon.social 3 hr. ago

@mitsuhiko @zzzeek @glyph I’m not entirely sure what we, as open source maintainers, are supposed to learn here, tho. To me, this is mainly a social problem: my AI policy is so liberal it gets regularly dunked on by LLM haters & yet none of the sloperators follow it because they just don't give a shit. They completely ignore any process or standards and just point their clankers at the bug track or idk. Nobody would call Mitchell an AI hater and yet his solution was shutting down contributions.

Paul Moore

@pfmoore@mastodon.social 2 hr. ago

@hynek @mitsuhiko @zzzeek @glyph We've found much the same on pip. We ask for responsible disclosure and respect for the maintainers' time, and mostly what we get is mountains of slop.

Framing that as an issue with open source maintainers rather than what it is - a social issue caused by people with no respect for the community who are being enabled by LLMs to cause harm - is a bad case of blaming the victims.

Armin Ronacher

@mitsuhiko@hachyderm.io 2 hr. ago

@pfmoore @hynek @zzzeek @glyph I just think we have not figured this out yet. It's not as easy as it looks.

~6 more replies

céline didone

@ruben@friendship.quest 2 hr. ago

@mitsuhiko@hachyderm.io @pfmoore@hachyderm.io @zzzeek@hachyderm.io Yeah but “we need time to learn” comes at a cost, one that comes at the costs of projects and their communities. Why is the burden on the maintainers and not the llm users to show they can be a part of these communities?

Armin Ronacher

@mitsuhiko@hachyderm.io 2 hr. ago

@ruben @pfmoore @zzzeek We always had to learn. Life is all about learning that has been true with Open Source since forever. Any new technology or changes required projects to adjust and coding agents are no exception.

céline didone

@ruben@friendship.quest 2 hr. ago

@mitsuhiko@hachyderm.io @ruben@hachyderm.io @zzzeek@hachyderm.io I’m not saying we don’t have to learn. AI Bans are a reaction to that system, it’s not slowing the learning process, it’s part of it.

But again, more energy directed at “well as a maintainer you should learn to deal with this”, and not enough into “if you want to use LLMs and be a part of this community, you have to show that you’re going to be a positive part of it instead of a burden/liability”

Both are learning processes.

Armin Ronacher

@mitsuhiko@hachyderm.io 2 hr. ago

@ruben@friendship.quest @ruben @zzzeek The thread here hands on Twisted's AI policy which even restricts AI contributions from core maintainers. That's a very different thing than closing contributions from untrusted individuals that might use AI.

céline didone

@ruben@friendship.quest 2 hr. ago

@mitsuhiko@hachyderm.io @zzzeek@hachyderm.io I can’t speak for why they did it, but it’s still the same. Core contributors aren’t isolated, they’re part of a system.

Use of LLMs excludes potential contributors that are against them eg. for ethical reasons. It’s as much a valid approach to say: “is the tool worth alienating this potential part of our community?”. Or, “will we nurture a more thoughtful practice by not leaning on this tool”. Plenty of potential reasons.

How that works out will be a learning. Maybe we find projects that do this attract more committed and intentional communities, maybe we find they don’t.

It’s as valid a part of the learning process as anything else.

Armin Ronacher

@mitsuhiko@hachyderm.io 2 hr. ago

@ruben @zzzeek Which is why I was curious how long that policy will stand. I find it harder and harder to imagine that an Open Source project can do that. For a start unless you use LLMs you are at a massive disadvantage when it comes to ensuring security.