Debian Votes on Eight Options for an LLM Usage Policy

The Debian project has put to a vote the question of how to treat contributions created with the help of large language models (LLMs). This concerns not only finished code, but any assistance from neural networks in work on the distribution.

The trigger was a proposal submitted in late July by Matthias Geiger. It called for an outright ban on any contributions to Debian created with the involvement or help of LLMs. The initiative sparked a heated discussion within the project and gave rise to a whole wave of alternative wordings.

In the end, developers will have to choose one of eight options. Their range is wide: from a strict ban on contributions prepared with the help of language models to wordings that, on the contrary, explicitly permit such a practice. In between these poles sit intermediate proposals.

A separate item is the option standard for Debian procedures, “none of the above.” If it wins, the project will be left without an agreed policy on LLMs — that is, the question will formally remain unresolved.

The very framing of the vote reflects just how ambivalently the free software community regards generative tools. For some participants, automatically generated code and text are a matter of quality, license cleanliness and responsibility for a contribution; for others, they are simply another tool that it makes no sense to ban. The vote in Debian shows that one of the oldest developer communities has no single answer yet, and that it has to settle the matter through a formal democratic procedure rather than by general consensus.

Source: LWN.net