Polish language model PLLuM is available
An AI assistant based on the PLLuM Polish language model is now available to Internet users. The model is designed to reflect the complexity of the Polish language, as well as to be free of harmful content and disinformation. NASK was part of the consortium that worked on the model, and is now taking on the role of project leader.
Developed as a result of a year’s work by the consortium on behalf of the Polish Ministry of Digital Affairs, the model is flexible and scalable – versions using between 8 billion and 70 billion parameters have been published. It is based on ethically sourced data – the training data has been used in accordance with copyright laws or their owners have licensed its use. A team of dozens of annotators worked on the quality of the answers provided by PLLuM. This, among other things, makes PLLuM a model that is safe and free of harmful content.
“A key role in the project is played by the verifiers. Thanks to their work, we know whether the model actually answers as we would like it to. We make sure it does not generate harmful content, or answer questions that we would not like it to answer, such as how to construct a bomb. NASK had a huge contribution here thanks to the experience of our experts. We specialize in cyber security, and we have experience working with harmful content. Our team is quite multidisciplinary. It includes programmers, lawyers, and people who have dealt with harmful content.” – explains Professor Szymon Łukasik, Director of the Artificial Intelligence Safety Research Center at NASK.
Thanks to its adaptation to the peculiarities of the Polish language, as well as the terminology of public administration, PLLuM will find wide application in the public sector. The project is entering the implementation phase, which will lead to the creation of an assistant for civil servants that will automate document processing, content analysis, information retrieval, and support in answering questions from citizens. Implementations for specific offices, tailored to their specific needs and characteristics, will be possible. The mObywatel Assistant, in turn, will support citizens in obtaining public information.
In this new phase of the project, NASK takes over as leader of the consortium, now called HIVE. In addition to NASK, the consortium continues to include Wrocław University of Science and Technology, Institute of Computer Science of the Polish Academy of Sciences, Institute of Slavic Studies of the Polish Academy of Sciences, National Information Processing Institute (OPI-PIB), and the University of Łódź. Two new partners are also joining the team: Center for Information Technology, implementing the model for civil servants and in mObywatel, and the Academic Computer Center CYFRONET AGH, which will provide its computing power.