Bielik Co-Creator: Poland Could Become the EU's Second AI Power

Remigiusz Kinas, a computer scientist and co-creator of the Bielik family of Polish language models, told Forbes Polska that the country has a real chance to become the European Union's second-strongest force in artificial intelligence after France. The catch, he says, is a decisive shift toward public investment, since Poland is currently losing ground to the world's most advanced research centers.

Kinas, a graduate of Poznan University of Technology and one of the creators of the Bielik project developed by the SpeakLeash foundation, doesn't hide the fact that Poland's resources today are far from those available to the biggest players in the AI market. Still, he believes that with the right mobilization of funding, computing power and data, a domestic team could close the gap far faster than Poland's current standing in global language model rankings would suggest.

A model in six months

According to Kinas, the key constraint isn't a lack of knowledge or team skill, but access to resources. He explains that with sufficient financial backing, computing power and access to high-quality data, including digitized Polish books, the team could build a language model on par with France's Mistral within six months.

In six months we'd build a Polish model on par with Mistral. And in a year, one competing with Chinese solutions - Remigiusz Kinas, Bielik co-creator

That's a bold claim, since Mistral AI's models rank among Europe's top open-source solutions and serve as a benchmark for many national AI projects across the continent. Kinas stresses that the scale of his ambition isn't detached from reality, but grows out of the experience gained through successive versions of Bielik, which has been developed for several years thanks to the community built around the SpeakLeash foundation.

The role of the Helios supercomputer

Bielik wouldn't exist in its current form without access to the computing power of the Helios supercomputer, launched by the Academic Computer Centre Cyfronet AGH in Krakow. That hardware made it possible to train successive versions of the model without relying on commercial infrastructure from foreign cloud providers, which would carry prohibitive costs for a non-profit project.

Kinas notes that dependence on a single supercomputer, shared with other scientific teams, is both a strength and a weakness of Poland's AI ecosystem. On one hand, it shows that competitive models can be built without the multi-billion-dollar budgets of American or Chinese giants. On the other, the scale of available resources still doesn't allow training models with parameter counts comparable to the leading systems from OpenAI, Anthropic or Google.

Poland's absence at ICLR

As an example of the widening scientific gap, Kinas points to this year's ICLR conference in Rio de Janeiro, one of the most important machine learning industry events, which drew roughly seven thousand researchers from around the world. In his view, the representation of Polish universities and research institutes was minimal, and the number of breakthrough papers originating from Poland remains disproportionately small relative to the country's talent pool.

That observation fits into a broader diagnosis Kinas has been making about Poland's AI sector for some time. The problem isn't a shortage of capable people, but a lack of institutional support that would let them stay in the country and publish results at the highest global level instead of emigrating to foreign research labs.

Poland's standing within the European Union

Kinas believes Poland has a real chance to become the EU's second AI power after France, but only if the government commits to serious investment and builds large research centers comparable to those in France or Germany. Without such a shift, he argues, the gap between Poland and the most advanced countries will keep widening, despite Bielik's growing popularity among companies and public institutions that use it as a free alternative to foreign providers' models.

For Polish businesses and public administration, this is a signal that using Bielik as a local, free language model doesn't answer the question of long-term strategy. If the state doesn't invest in computing infrastructure and research programs, future versions of Bielik may find it increasingly hard to keep pace with foreign models, even with the best will and skill from the SpeakLeash team.

Sources: Bielik co-creator on Polish AI (forbes.pl), How Poland can catch up with AI leaders (wnp.pl)