A couple of weeks ago at a development meeting we settled on a change to the code of conduct regarding our stance on AI.
Here is it!
Expected Behavior
…
Usage of AI tools
The community values the thoughts and ideas of others, using an AI to develop one’s thoughts is acceptable, but posting the output of an AI ‘as if’ it came from a human is not.
AI slop and AI based low effort contributions are not tolerated and will be judged by moderators discretion on a case by case basis.
Do not use any kind of autonomous bots (i.e. OpenClaw etc.) for any kind of interaction with the community or development!
Ensure all AI‑generated text or code has been carefully reviewed and edited by yourself.
Do not paste other user’s content or code into LLMs without their consent.
AI tools, such as translators, may be used to help one participate and contribute to the community, but they should be used in moderation, or as a last resort. The community would rather engage with human imperfection than a confident machine.
…
Essentially: using AI as a tool is fine, using AI to generate code must be declared and you should put effort into understanding what it produces (more applicable to PR’s over on GitHub than here), and it is not acceptable to present a chat bot as a human. This reflects the environment and culture we as developers wish to participate in.
These changes are currently only visible over on GitHub and don’t apply to this forum, I don’t know if one of the @moderators wants to adopt it?
Would it make sense to move supercollider off github then? I guess that platform is all about putting other people’s code into LLMs without consent, and afaik you have to interact with an autonomous bot to use it.
I’d certainly would like to get off of GH, but we require a zoo of machines for building/testing, i.e. windows/linux/macOS [intel and arm] - GH is giving this us for free, and we would either have to pay someone and/or someone would need to come up w/ a custom setup. But we don’t have any kind of funds and I think time can probably spend better developing SC rather than a CI, but thats debatable.
Codeberg seems to build on forgejo actions and woodpecker. Forgejo is linux only (therefore not suitable) and woodpecker runs everything in a docker virtualization IIRC, which is not really suitable for RT sensitive test suite.
We could ask e.g. if IEM Graz could provide us some infrastructure, but I am also reluctant to tie everything to this university - they already host pd.
I personally would also prefer a self-hosted gitlab over a codeberg b/c I much prefer the gitlab UI over the GH one.
Last time I checked we would have to pay 100$+ for our own CI, so this seems not really feasible. Would be great if woodpecker would not rely on containers and make a business out of this so we can pay-as-we-go, but they are not offering this - though gitlab offers this.
So one move could be towards a non-self-hosted gitlab, but I also don’t like their approach towards AI, so I doubt that much is gained there. See e.g. their open source program: Community programs | GitLab Docs
We could also move issues to this forum, though reviewing of PRs would still need to happen on GH, so I think we introduce more noise w/o much effect [e.g. usernames don’t align]. I don’t have the illusion that code on codeberg, content of this forum etc. isn’t scraped and ingested into AI training after all.
huh? dare to share more about this? I personally have opt-out of all co-pilot stuff, so I don’t see anything
I can see the dilemma, but I think it should still be recognised that the code of conduct is being broken, as all the supercollider code is being hosted by an extractive microsoft AI platform.
Fair enough, I am probably wrong about that, I don’t really use github myself.