would any fedi ppl wanna get together to brainstorm a code of conduct and contribution policy for the current times
feel free to also just suggest points that should be in it and your own experiences
0 Boosts
0 Likes
This federated comment is not yet available here.
@SuperDicq @lumi have a formal document to point to when making a moderation decision is extremely helpful to avoid drama and accusations of unfairness
@navi @SuperDicq @lumi Yeah, CoCs are useful as a "Don't make me point the sign" and having one to point to.
@lanodan @SuperDicq@minidisc.tokyo @lumi @navi I personally like the "yeah, it's unfair and arbitrary, this is not a democracy, if you don't like it the door is right there" response.
Of course it means the project relies on the dictator's good ethics entirely and that is not applicable to larger projects, does not survive maintainer change, etc. I'm not saying I believe in it. But it is satisfying to bully assholes who would bully you first chance they get.
This federated comment is not yet available here.
@SuperDicq @mirabilos @navi And so I guess in like 10~20 years given how deeply screwed the hardware market is.
This federated comment is not yet available here.
@SuperDicq @lanodan @mirabilos
the majority of binaries i use were compiled in my system, with some in gentoo's donated hw servers =w=
everything i heard about "from scratch" llms points to "they're nowhere even close to the ones people use", and still take massive gpu compute most people don't have; people are not going to be used locally trained locally ran llms for the stuff we're discussing anytime soon, because they're just not going to generate "acceptable" looking output
the majority of binaries i use were compiled in my system, with some in gentoo's donated hw servers =w=
everything i heard about "from scratch" llms points to "they're nowhere even close to the ones people use", and still take massive gpu compute most people don't have; people are not going to be used locally trained locally ran llms for the stuff we're discussing anytime soon, because they're just not going to generate "acceptable" looking output
This federated comment is not yet available here.
@SuperDicq @mirabilos @navi @lumi Except the training doesn't grows on trees, specially when it visibly* needs to be done regularly as otherwise it ends up printing out obsolete data (like old APIs, or sometimes even old known-broken code…).
*From what I've seen from slop PRs
*From what I've seen from slop PRs
@SuperDicq @lanodan @mirabilos
those llms had to be trained, in massive data centers, scraping the whole internet; that step burns down the planet and with current models it's unavoidable
those llms had to be trained, in massive data centers, scraping the whole internet; that step burns down the planet and with current models it's unavoidable
This federated comment is not yet available here.
@SuperDicq @navi @lumi Yeah, and such redaction means it pretty much needs to be project-controlled, rather than some random third-party software.
This federated comment is not yet available here.
@SuperDicq @icedquinn @navi @lumi So far there's not even the slightest hint such an free and ethical LLM could even exists, and policies aren't written in stone so exceptions could be made, and I'd say it's better to be deny-by-default on things that have the risk of tainting projects (both copyright-wise and community-wise).
And like Gentoo Policy even has explicit text about revisiting things if such a thing comes up.
And like Gentoo Policy even has explicit text about revisiting things if such a thing comes up.
This federated comment is not yet available here.
@SuperDicq @navi @lumi
> Privacy doesn't really apply as software discussions tend to be public.
Yeah, although not always the case (say for channels dedicated to security issues, and sometimes bits of bureaucratic stuff too) and things like having public IRC logs is far from unanimous.
At least typically I'd refuse to have public logs if it's a channel with user support, too easy for one to unintentionally leak something they're rather keep slightly private / off-the-record.
> Privacy doesn't really apply as software discussions tend to be public.
Yeah, although not always the case (say for channels dedicated to security issues, and sometimes bits of bureaucratic stuff too) and things like having public IRC logs is far from unanimous.
At least typically I'd refuse to have public logs if it's a channel with user support, too easy for one to unintentionally leak something they're rather keep slightly private / off-the-record.
This federated comment is not yet available here.
@SuperDicq @navi @lumi Yeah due to:
- Authorship concerns as none of the LLMs respect even permissive licensing
- Privacy and spam concerns if they plug it into discussion spaces
- Authorship concerns as none of the LLMs respect even permissive licensing
- Privacy and spam concerns if they plug it into discussion spaces
@SuperDicq @lumi @navi And to be clear, none of my policies are specific to LLMs, an IRC bot written in ancient Perl would be subject to the exact same rules.
so it will include the usual coc stuff but also:
- full genai ban for creation
- heavily discourage and condemn genai for everything else
- newer / untrusted contributors should talk about their methodology or get prior approval
- hard anti-fascist stance
- contributions not done in good faith can be closed immediately
maybe other stuff? would love to hear from yall
@lumi cat ears recommended. If people can't handle cat ears they're going to be bad for community
