Cross-Border Scientific Cooperation in the Development of Polish LLMs
A technical workshop was held at NASK headquarters, bringing together the HIVE AI consortium and representatives of the French startup Mistral AI, known for its expertise in open large language models (LLMs).
The meeting focused on exchanging knowledge and developing strategies for multilingual model adaptation and training, with particular emphasis on optimizing computational efficiency and data quality.
Under the leadership of Dr. Agnieszka Karlińska (NASK) and Dr. hab. Piotr Pęzik, Prof. of the University of Łódź, the HIVE AI team presented proprietary instruction and preference datasets and discussed the requirements for pretraining LLMs from scratch. The workshop resulted in a framework for continued collaboration aimed at strengthening sovereign AI capabilities in Poland.
Founded in 2023 by former engineers from Google DeepMind and Meta, Mistral AI promotes openness and transparency in AI development by releasing its models under open-source licenses. Within a short time, the company has become one of Europe’s most innovative independent AI labs.
Participants included Benjamin Trom, Szymon Antoniak, and Jan Ludziejewski from Mistral AI, alongside nearly 20 Polish researchers representing institutions such as NASK, Wrocław University of Science and Technology, University of Łódź, the National Information Processing Institute, Academic Computer Centre Cyfronet, and the Institute of Computer Science of the Polish Academy of Sciences. The event underscored the importance of international cooperation in building responsible and innovative AI solutions.