SafetyTransparencyAccountabilitycs.CL
PLLuM: A Family of Polish Large Language Models
Jan Kocoń, Maciej Piasecki, Arkadiusz Janz, Teddy Ferdinan, Łukasz Radliński, Bartłomiej Koptyra, Marcin Oleksy, Stanisław Woźniak, Paweł Walkowiak, Konrad Wojtasik, Julia Moska, Tomasz Naskręt, Bartosz Walkowiak, Mateusz Gniewkowski, Kamil Szyc, Dawid Motyka, Dawid Banach, Jonatan Dalasiński, Ewa Rudnicka, Bartłomiej Alberski, Tomasz Walkowiak, Aleksander Szczęsny, Maciej Markiewicz, Tomasz Bernaś, Hubert Mazur, Kamil Żyta, Mateusz Tykierko, Grzegorz Chodak, Tomasz Kajdanowicz, Przemysław Kazienko, Agnieszka Karlińska, Karolina Seweryn, Anna Kołos, Maciej Chrabąszcz, Katarzyna Lorenc, Aleksandra Krasnodębska, Artur Wilczek, Katarzyna Dziewulska, Paula Betscher, Zofia Cieślińska, Katarzyna Kowol, Daria Mikoś, Maciej Trzciński, Dawid Krutul, Marek Kozłowski, Sławomir Dadas, Rafał Poświata, Michał Perełkiewicz, Małgorzata Grębowiec, Maciej Kazuła, Marcin Białas, Roman Roszko, Danuta Roszko, Jurgita Vaičenonienė, Andrius Utka, Paweł Levchuk, Paweł Kowalski, Irena Prawdzic-Jankowska, Maciej Ogrodniczuk, Monika Borys, Anna Bulińska, Wiktoria Gumienna, Witold Kieraś, Dorota Komosińska, Katarzyna Krasnowska-Kieraś, Łukasz Kobyliński, Martyna Lewandowska, Marek Łaziński, Mikołaj Łątkowski, Dawid Mastalerz, Beata Milewicz, Agnieszka Anna Mykowiecka, Angelika Peljak-Łapińska, Sandra Penno, Zuzanna Przybysz, Michał Rudolf, Piotr Rybak, Karolina Saputa, Aleksandra Tomaszewska, Aleksander Wawer, Marcin Woliński, Joanna Wołoszyn, Alina Wróblewska, Bartosz Żuk, Filip Żarnecki, Konrad Kaczyński, Anna Cichosz, Zuzanna Deckert, Monika Garnys, Izabela Grabarczyk, Wojciech Janowski, Sylwia Karasińska, Aleksandra Kujawiak, Piotr Misztela, Maria Szymańska, Karolina Walkusz, Igor Siek, Jakub Kwiatkowski, Piotr Pęzik
November 5, 2025
arXiv.org
13 citations
Key Takeaways
-
Language diversity matters for responsible AI: PLLuM demonstrates that English-centric LLM development creates gaps in AI access and quality for non-English speakers. By building models specifically for Polish, the consortium ensures that responsible AI practices—including safety filtering and cultural relevance—can be properly implemented for underrepresented languages rather than relying on translations or adapted English models.
-
Transparency and governance are foundational: The paper emphasizes a "Responsible AI framework" incorporating "strict data governance" as a key component, suggesting that responsible AI development requires deliberate structural choices about data sourcing and management from the outset, not as an afterthought.
-
Safety requires multiple layers: PLLuM implements a "hybrid module for output correction and safety filtering," indicating that responsible AI systems benefit from combining different safety approaches rather than relying on a single mechanism to prevent harmful outputs.
-
Open-source development supports accountability: By releasing PLLuM publicly rather than restricting it to commercial use, the consortium enables external researchers and the public to audit, test, and verify the models' behavior—a critical component of responsible AI that proprietary systems cannot provide.
-
Culturally-informed AI is more responsible: The development of a custom 77k instructions dataset and focus on "culturally relevant" models suggests that responsible AI must account for local contexts, values, and needs rather than assuming one-size-fits-all solutions work across different societies.