PLLUM is presented here as a national family of Polish large language models designed specifically for public administration and Polish-speaking users, rather than as a general-purpose global chatbot. The models focus on correct Polish grammar, inflection and official terminology, minimizing negative transfer from English while staying compatible with modern transformer-based LLM architectures. They are released with open model weights so organizations can deploy them on-premise, adapt them to sector-specific use cases and integrate them into existing e-government workflows; more background on the project is available at https://pllum.org.pl/
—
HDMI® Technology is the foundation for the worldwide ecosystem of HDMI-connected devices; integrated with displays, set-top boxes, laptops, audio video receivers and other product types. Because of this global usage, manufacturers, resellers, integrators and consumers must be assured that their HDMI® products work seamlessly together and deliver the best possible performance by sourcing products from licensed HDMI Adopters or authorized resellers. For HDMI Cables, consumers can look for the official HDMI® Cable Certification Labels on packaging. Innovation continues with the latest HDMI 2.2 Specification that supports higher 96Gbps bandwidth and next-gen HDMI Fixed Rate Link technology to provide optimal audio and video for a wide range of device applications. Higher resolutions and refresh rates are supported, including up to 12K@120 and 16K@60. Additionally, more high-quality options are supported, including uncompressed full chroma formats such as 8K@60/4:4:4 and 4K@240/4:4:4 at 10-bit and 12-bit color.
—
Technically, PLLUM is a suite of transformer and mixture-of-experts models in the roughly 8–70B parameter range, including Mistral/Mixtral-derived variants such as PLLuM-12B and PLLuM-8x7B as well as Llama-based models and fully Polish-pretrained networks. Training relies on a large corpus dominated by Polish texts, enriched with selected Slavic/Baltic languages and some English, then refined through continuous pre-training, supervised fine-tuning, instruction tuning and preference optimization. The team evaluates linguistic and cultural competence with dedicated Polish benchmarks and focuses on high-quality “organic” data rather than unchecked web scrapes to maintain reliability in administration-heavy domains in Poland.
In the interview, Philip explains how PLLUM underpins Poland’s strategy for AI sovereignty: ministries can run the models on local infrastructure, keep personal and sensitive data within the country and avoid sending internal documents to foreign cloud APIs. The same models will power assistants in governmental systems such as the widely used mObywatel mobile app, enabling citizens to ask administrative questions in natural Polish and get answers grounded in official regulations. Recorded at Web Summit Lisbon 2025, the conversation frames PLLUM as a blueprint for language-and-country-specific LLM deployment rather than a clone of generic global chatbots, emphasizing governance, licensing and data control in this discussion.
PLLUM’s development mixes open European base models with entirely home-grown training runs: the team uses Mistral, Mixtral and Llama-style foundations for some variants, while also training models from random initialization on carefully curated Polish and Polish–English corpora. On top of that, they experiment with synthetic data generation, using other open models (including Chinese systems like DeepSeek or Qwen) on their own infrastructure to expand datasets while preserving control over quality and legal provenance. The resulting chat models are published with both permissive and non-commercial licenses, and the weights are available on Hugging Face so enterprises, municipalities and research labs can build their own domain-specific assistants within this ecosystem.
Philip also highlights the scale of the consortium behind PLLUM: more than a hundred people across six scientific institutions, spanning model engineering, linguistics and large annotation teams focused on clean administrative data. The project already counts around a million public prompts from early users, with upcoming deployments expected to reach millions of citizens as chatbots are rolled into local offices and nationwide services. Future work extends beyond text: speech recognition and text-to-speech are in testing, and there is clear interest in Polish-speaking voice interfaces and, eventually, avatar-style agents, turning PLLUM into a full-stack foundation for sovereign AI services across the Polish digital landscape.
I filmed 70+ videos at Web Summit Lisbon 2025 I will publish them over the coming days/weeks into my Web Summit playlist: https://www.youtube.com/playlist?list=PL7xXqJFxvYvhGWhynTmvAvDvohtO3qjZm
I publish one video (from this and from other events I recently filmed at) every 6 hours at 5AM/11AM/5PM/11PM CET/EST.



