There is an interesting discussion going on in the OSMF run discourse instance on community.osm.org regarding regulation of use of LLMs in communcation there.

First a bit of background

The OSMF in 2021 introduced a new behavior control framework for the forum and some mailing lists – which i criticized for its vagueness (and resulting subjectivity in implementation – giving space to cultural bias) and its intolerance towards behavior diversity, both culturally and as a result of neurodiversity.

Later in 2025 the OSMF introduced an AI policy supplement to those behavior rules which essentially extents the vagueness of the original framework to the field of use of LLMs.

These policy documents and their practical application reflect the idea of behavior control as a system of authority where the decision on what behavior is considered acceptable is determined centrally by a group of people (the moderators) and the users of the channels need to trust that these perform their work in a well meaning and competent way (which in my eyes is mostly the case) and universally avoid their cultural and personal predisposition creating an unfair bias in how they judge different styles of communication.

Within the context of the discourse forum the idea to trust the integrity and objectivity of the moderators also stands in conflict with the technocratic paradigm of the underlying software, which is based on a strong faith in technology and firmly delegates all kinds of social functions to technical automatisms. This manifests most clearly in communication participants being structured in an social hierarchy defined through some intransparent scoring. This practically also means communication is sometimes hidden from users without this being transparently communicated to either the moderators or users.

This technocratic management is partly defined by the developers of the software the OSMF uses, partly by the self appointed technical management of the forum – which works completely separately from the official moderation.

Many of these things i have individually criticized in the past already, but i thought it is important for the matter at hand to summarize them together.

On the matter of technically generated communication

I won’t bore you with yet another of those my personal view on AI use comments you can find endlessly around the internet. If you are interested in some perspective on this from me i can point you to what i wrote about this in the context of mapping many years ago – which essentially still applies today.

What i want to comment on here is that the whole issue with using LLMs in discussion is a problem to a large extent resulting from the focus on the two described methods of regulating group communication – the moderator authority you need to trust and the technocratic management based on faith in technological management of social relations on a collective level. The combined application of these two principles has deprived people on the forum of both the technical ability and the practical experience with managing their social interactions on that platform themselves.

I probably need to explain this a bit further. Back in the days of Usenet one of the most significant and defining elements for the social interaction in these groups was the feature of having a so called killfile in practically all NNTP readers used by serious participants in those groups.

A killfile represents the ability of the individual communication participant to specify criteria for communication they want to not read, most commonly communication from certain participants. See the Jargon File for details. While this feature was rarely really used the fact that it was readily available to nearly everyone participating in those group communications on Usenet had a substantial effect on the communication culture. I will go out on a limb here and claim that if Usenet still was widely used these days use of LLMs in Usenet communication by human participants (in contrast to LLM generated spam – which is a completely different matter) would not be a problem.

Bottom line: I think the whole problem of LLM use in communication on the OSMF forum in the form discussed under the link on top is the result of the ill-advised attempt to collectively manage group communication of the participants (telling people what of the communication to read and how to read it, what to write and how to write it) instead of giving them the ability and the freedom to manage that for themselves individually.