In the third and final panel of an unplanned triptych of post-AI thinkpieces (see previously, what happens and what is post-AI), I’m sharing the free/open source Strudel project’s new post-AI policy.
Community
All community members should be treated with respect and kindness. We welcome contributors to the community who are ready to follow this policy as well as the above code of conduct.
- We wish to protect the humanity of our communities and projects, to support the motivation and well-being of all of its contributors and users.
- To this end, contributors and other users of our projects should not have to engage with LLM-generated code and text as part of their community work.
- We recognise that LLM-based scraping and generation of (A)GPL code shared by ourselves and others is of questionable legality. For example, under many juristictions, LLM-generated work does not have a copyright holder and can therefore not be licensed under the (A)GPL.
- A contributor’s personal use of LLMs and related technology is their own business, whether for research, analysis, bug discovery or other assistive technology, so long as the output is not included in any contributions to our projects.
- Contributions including LLM-generated content will be declined.
- LLM-generated text (e.g. commit messages, issue comments, forum posts etc) generated by LLMs will be flagged, and ultimately deleted.
- We encourage asking each other questions and working through problems to promote collective knowledge and creative ideas (never telling each other to “just google it”).
- We strongly discourage others from forking/importing our projects into corporate AI platforms, including Microsoft Github.
- All contributions must be submitted by a person who understands the changes, is happy to answer questions about them, and where possible, help maintain the code in the future.
- In community discussion channels (at the time of writing, this refers to the uzu discord server), all discussion (including criticism) of LLM technology and use should be kept to designated channels.
The latest version can be found here.
It’s taken GCC’s policy as a kind of starting point, but has a broader focus on community, motivation and well-being. It avoids trying to police what people do in the comfort of their own laptops, not making a value judgment on using LLMs, beyond saying LLM output shouldn’t reach the project’s repository as code, comments, issues or anything else.
This preserves strudel as a community, in that maintainers and other contributors can volunteer their time to work on the project on the presumption that they are interacting with other people, and not taking part in some weird dystopian ritual with bots. It also gives peace of mind to end-user live coders who come to strudel to enjoy live coding as a human expression, in an environment crafted by other people, as a holiday from the current take-over of our online and work lives.
It doesn’t ‘other’ people, dividing people who are touched by ‘slop’ and those who are not. In reality many of the more vocal people defending live coding against an AI takeover do use LLMs as part of their day jobs, perhaps under duress, and these tools are by now used very widely, including whenever you do a google search.
However, the threat of an AI takeover of live coding is real – from the beginning, live coding was always a human-centric movement against automation and generative art, but Anthropic have approached Strudel wanting to donate in return for interoperability, as well as looking for Strudel artists willing to take five figure sponsorship. To avoid becoming a marketing wing of a corporation, we need to form productive coalitions to continue live coding as humane culture, connecting with related counter movements.
This also of course preserves Strudel as a code base, just in case the legal uncertainty around LLM use in free/open source software results in LLM output being un-GPLable. This is also a threat real enough, hence the gcc policy. However I remember the response to my questions around authorship of generative music output, when I was first getting into algorithmic music in the early 2000’s while working in the independent music industry. The response to a record label music rights person was an evasive ‘follow the money’. The implication is that if there’s serious money to be made, then the legal infrastructure would bend/grow accordingly. In my case there wasn’t, and so it wasn’t a question worth answering.
I called it a live coding policy for software development, because I think live coding is a special case. Live coding is something people do to make music and art, and to automate it changes it into an entirely different category of activity. There is no clear line between live coding the code that makes music, and coding the environment for live coding – it’s an arbitrary line that people draw in different places depending on on their interests. I’ve long thought live coding is a totally different activity to software engineering, and honestly hope they continue to grow apart, with live coders taking their language makers with them.
So this ended up as a mild and pragmatic policy so far (it might change), but I think also radical in its own way, for focusing on human interaction, banning all LLM text in those interactions, as well as directing people away from the sickness that is Github. This makes implicit that Strudel is a playground for free, collective thinking and innovation around time and rhythm, and not competitive development of throwaway products at speed. There are by now hundreds if not thousands of vibe-coded apps using strudel, probably with an average of less than one user each. A free/open source project like Strudel is not just software, but a community of practice.
Maybe I’ll start putting something in my email signature, like “Please do not use LLMs in replies (beyond e.g. translation and assistive use), scrappy messy language welcome.” As the ketamine billionaire dystopia grows around us, I’d like to at least keep my interactions with people humane.