Umfeld×LLM 

this is a draft for a statement regarding the development of Umfeld and the use of LLMs in the process. it is also a reaction to a recent development on Codeberg 1 2, the platform that is also hosting the Umfeld project, banning “projects that mostly consist of code written by ‘generative AI’-tools” 3.

while most of the Umfeld project IS written by humans ( mainly myself ), there are components that are ( deliberately ) generated by LLMs, i.e, this move affects the Umfeld project.

while I mostly understand the motivation behind the ban, I do not agree with it in its current form. to me, it feels like the people at Codeberg identify the right signals but implement the wrong solution. a lot of things are wrong about the economical and ecological aspects of LLMs ( and other so-called AI as well ). many other aspects, psychological or social, are still vague or not yet properly understood or researched. however, banning ( or blocking or, if you will, cancelling ) this technology at this point just does not resonate with the way i ( or dare i say we ) usually deal with things. therefore, i want to propose a different approach, that is more inclusive, discursive, and ( IMHO ) productive.

i borrow the legal term fair use to frame my proposal, in order to suggest that it is a) about use and b) about fairness in multiple directions, however, not about copyright but rather about doing something that is somewhat fair i.e justified or justifiable. this is a first draft:

Umfeld / Fair LLM Use Statement

when developing Umfeld, developers may choose to make use of coding agents and LLMs. however, the usage is constrained by a Fair LLM Use policy. the fair in this use case is constituted by the following points:

  1. developers commit to becoming better at producing hand-written code.
  2. developers always and actively contemplate the implications of LLMs in regard to political, ecological, psychological, technical, and social effects.
  3. developers actively introspect themselves in regard to the effect that LLM usage has on themselves and on those around them. they commit to debate, collectively reflect, and care for each other in this regard (and in general).
  4. developers try to find a balance between hand-written code and code generated by LLMs. they will also find words to argue for their definition of balance.
  5. developers remain ( and feel ) forever responsible for the code they instructed an LLM to generate. all code contributions are signed off by a human being.
  6. developers predominantly use LLMs in cases where they would otherwise struggle significantly to complete a task, e.g:
    • when using libraries with poor documentation or examples.
    • when using programming languages they do not know well ( yet ).
    • as a temporary replacement for a human developer that unexpectedly vanished.
  7. developers may critically and playfully explore LLMs and coding agents, as they so often do and did with new technologies.

to be continued/developed …

@edit(dpp, “renamed LLM Fair Use to Fair LLM Use, extended comment, changed URL”)

  1. Protecting our FLOSS commons from LLMs ↩

  2. Locking LLM decisions related posts lacking constructive criticism #190 ↩

  3. Codeberg’s Terms of Use (TOC) § 2 (1) 7 Allowed Content & Usage ‘You must not share projects that mostly consist of code written by “generative AI”-tools …’ ↩