First, as a disclaimer, I am not a DD or officially part of the Debian project, and did not participate into any vote. I helped with release testing from time to time, did support online for other users, but that's it.
So, as you may imagine, this is about the recent vote most of you participated in. Every tool you (as a community) made, dpkg, apt, (cross)-compiler infrastructure, mirrors, etc. You made it with your own means and resources, as a community first, and as an individual second. The community made it possible to get where the project is today (and I'm very proud to be a Debian user) Modern large language models (not the concept of a language model in itself, but the "product"), on the other hand, are only made possible by vast amounts of capital, energy and infrastructure expenditures. These will always be needed to train and build a model, be it closed (like all iterations of Claude) or open-weight. These are very high costs. >From the outside, I've also witnessed how much of training also carries its own toll on the FOSS movement as a whole. This is a cost too, that is passed onto the community, without its consent. For example, agressive scraping leading to DoS. Even the linux kernel project has had to put up guards to keep their git available for themselves. They also had to deploy LLM-against-LLMs, to deal with the enormous amounts of LLM-assisted (wisely or not) patches being sent. I'll stop there because I guess, most of you know the story. Comparing to large language models "as a tool", do we know any other community projects where the costs for their development were this high? I cannot think of any, personnaly. If someone did a tenth of that damage to another FOSS project, they would be rightfully banned. Anyways, my personal feeling, as a long-term Debian user since 10 years, is that I am surprised of this decision. Debian is known to be conservative and cautious about novelty, which is a quality in itself. To say that the project is fine with allowing itself to become, as a whole, potentally more reliant on something (LLM models) that: a) It cannot redo by itself, even with all of its resources b) Which needs enormous and continuous capital expenditure to stay relevant and useful c) Which we don't know if it they will continue to be available under the same conditions in the future, or at all d) Which leads to higher costs (both expenditures and social) on the project itself, and on upstreams Is hell of a bet on the future, and on its future, to make. As a long time user, I am sincerely worried about the direction that Debian is currently taking. BR
