Krusader's Talk

What do European strategic autonomy, local language model projects, and a puppy have in common?

In the latest episode of the Danish National Radio podcast Udsyn, Kaspar Colling Nielsen and Marlene Wind, Professor & Director of the Centre for European Politics at the University of Copenhagen offer a fresh and highly relevant perspective on Europe’s “little brother” mindset toward the United States and on its own role and responsibility in regional security policy, and present the idea that the new American discourse on military and trade alliances could be a positive opportunity to “grow up” again; one that we Europeans should seize.

Let’s go MEGA!

The same opportunity is equally open for digital Europe. Right before our eyes, generative AI is catalyzing a broad transformation of Europe’s digital landscape, ushering in a new technological paradigm.

Within this transformation lies the option of achieving technological independence. Using large language models, we can grant citizens and businesses digital freedom, as language itself provides the input necessary for machines to act on our behalf. Ironically, though, this freedom is tightly bound to its source (the LLM) that links our words to digital capabilities. So it makes sense to keep control of this critical component of future digital infrastructure, doesn’t it? Or should we keep relying, as we largely do today, on American tech giants (like Microsoft) and accept that it comes with tough commercial conditions — ones that might be more sensitive to the strategic interests of the U.S. government than our own?

Admittedly, that was a long, rhetorical question. Because of course, the future digital Europe, including Denmark, should have European AI under the hood. In this spirit, it is great news that the OpenEuroLLM project has secured €20.6 million in funding from the Digital Europe Programme under the EU Commission. However, a small disappointment is that no Danish stakeholders are part of the project. On the other hand, we can take joy in the parallel Danish Foundation Models research project, which, with substantial funding (DKK 30.7 million) from the Danish Digitization Agency, is investigating and developing AI for specific Danish use cases. It is not immediately clear how strategic collaboration happens between these two projects (and others). But the great thing is that models can always be refined by integrating data from other projects.

Kruso, too, is amidst several AI projects that rely heavily on processing linguistic nuances for Danish, German, and Dutch clients, and we’re getting a bit impatient. It’s tempting to bypass, for example, Microsoft and OpenAI from the outset. Yet we must acknowledge that right now—especially for public organizations—it often makes sense to experiment with AI accelerated by the many tools in the Azure ecosystem. That’s certainly more productive than sitting idle, and any insights gained can likely be carried over, as long as we’re meticulous about designing a model-agnostic AI architecture.

In many ways, I identify with Palle, my 8-month-old puppy. At first glance, he looks harmless. But he is impatient. The picture fails to convey that ten minutes later—frustrated that I am stuck in a Teams meeting—he will vent his impatience by digging in a flower pot, where perhaps a deeper meaning can be found. In the same spirit, I must also come to terms with the fact that mature, shared European AI stacks won’t be available for my projects anytime soon. Instead, I will have to contribute by digging up value in smaller, independent projects here and now. And, of course, I can also encourage everyone to consider whether they can contribute by open-sourcing datasets for the maturation of Denmark’s future language model – https://alexandra.dk/dansk-sprogmodel-konsortium/

– Let’s go MEGA!