10 Post
Feed post
IN THE NEWS: Switzerland just did what everyone’s been talking about.A national, open-source language model built for the public good.I’ve been having conversations about a publicly funded, open-source language model for a while now. A model designed not just for scale or market share but for trust, transparency, and public value.And now the Swiss have done it.ETH Zurich, EPFL and CSCS have built and trained a multilingual LLM on a national supercomputer, using a dataset spanning over 1,500 languages. It’s open source, Apache licensed, and aligned with both Swiss data law and the EU AI Act. They’re calling it “a language model built for the public good”.It’s not trying to out-market anyone. It’s trying to set a different example one rooted in research, education, and civic purpose. With transparency in training data. With ethical oversight. With intention.Is this the future?Should more countries be developing their own models built on public infrastructure, governed by public values?Because if we want to trust these systems, shouldn’t we also trust how they came to be? Who they’re accountable to? What they’ve been optimised for?Maybe this is what it looks like when AI isn’t just another product.Still sitting with this.https://lnkd.in/eHbhRQYxhashtagResponsibleAI hashtagOpenSourceAI hashtagAIForPublicGood hashtagDigitalSovereignty hashtagEthicalAIDevelopment*note all Tshirt images are AI generated and not real or in anyway purchasable.
Reactions
- [
- [
- [
- [
- [
- [
- [
- [
10 Replies on Alan Robertson’s comment
9 Replies on Alan Robertson’s comment
The phrase, “Governed by public values,” has Newspeak written all over it.
9 Replies on Christopher G.’s comment
Christopher G. The idea that competition is the only driver of innovation is such a stupid concept. People innovate for so many damn reasons. Because they're passionate, because they want to advance things, because they like creating, because it's fun, because it, because they wanna challenge themselves, etc, etc, etc. You don't need competition to drive innovation. That's such a limited worldview.If competition for resources or markets were the only drivers of innovation, fundamental science and pure math wouldn't be a thing. Art wouldn't be a thing. Music wouldn't have developed. People innovate because that's what humans fucking do all the time.
Happy to see that my company Nebuli is not so lonely anymore in the "No VC" front. Many congrats to Switzerland. We need more teams and organisations that push for ethical AI and innovation. Side note: I find it astonishing that the UK is not positioning itself to be part of this conversation. Outragous!
3 Replies on Tim El-Sheikh’s comment
https://www.linkedin.com/in/timelsheikh/Tim El-Sheikh UK? When was the last time a publicly funded project succeeded in the UK? (i.e. delivered on time and money with clear added value) You need a competent government and civil service for that, something that has been missing in the UK for generations.
I think you are misunderstanding the terms....an LLM eployed on public infra isn't open source..an LLM to be open source needs the training data open sourced too and be capable to run in local
6 Replies on Carlo D.’s comment
We will see... If not, not much of a difference
3 Replies on Tamara Polajnar, PhD’s comment
This move feels like a real, tangible step in the direction of building AI that’s actually concerned with public values, not just private ambition. It will be interesting to see if others follow this direction. I saw a similar ish project that the Quebec AI Institute and the University of Toronto have done, to create open-source models focused on social good which is very interesting too.
1 Comment on Rebecca Harper’s comment
https://www.linkedin.com/in/rebecca-harper-15054018/Rebecca Harper It certainly does, I believe Poland does the same thing
Good post. Yes, I’ve been thinking about this as well for a long time. We should really have a national LLM project just like we have a Library of Congress. How these LOM’s are trained and the training data that goes into it is crucial. It is a public resource just like a library just like universities.
2 Replies on Mark Hodgson, M.Sc.’s comment
Interesting, but training model on public source code is not sufficient. Every other model also has been trained, but code generation results are very poor still in 2025. Someone has to write code generation application thorougly. There can't be shortcuts like training. The rules coded must be hardcoded exactly how compiler would treat the code.
3 Replies on Vadims Kurmis’ comment
https://www.linkedin.com/in/gergely-vandor-49989715/Gergely Vandor I mean whatever solution fits best, but accuracy is big problem. It could be hardcoded rule based generation in additon to improved models which should discard low quality data from training on. I'm not expert in AI, but what I think is that just training is not enough. The model itself must process data as compiler would process source code. Someone has to spend what it takes and go through language specification, and make AI know the boundaries. In highly competing market someone has to improve accuracy. What we have now - can't be used. For autocompletion and prototyping - maybe, for actual code generation - not yet, still too low on accuracy.
Thank you for sharing. I’m not as optimistic when it comes to building national LLMs. Switzerland clearly had the resources and infrastructure to make it happen, but not many countries are in that position. Most can’t realistically keep up.
6 Replies on Milica Jeremic’s comment
https://www.linkedin.com/in/milica-jeremic/Milica Jeremic rather than looking at them as national LLM, I look at these from the lens of publicly funded LLMs not owned by a private entity. There is a whole issue of AI sovereignity. Why should countries cede it to transnational private entities. There needs to be an alternative available. LLMs today are like shrinkwrap software of yesteryears. Open source can’t develop foundational models due to ultra high costs. Privately donated open source foundational models (like Llama) are at the mercy of private deep pockets.