Kyivstar, Ministry of Digital Transformation sign memo of cooperation to create LLM
The largest Ukrainian telecommunications operator Kyivstar and the Ministry of Digital Transformation signed a memorandum of cooperation on a large language model (LLM).
"We will be able to launch services in the country without storing data abroad or in other places that are questionable or may be affected. We will store this on our own infrastructure, using our own LLM material. Perhaps not only it, but fundamentally this model, as well as other products. And this will allow us to launch more services," said Deputy Prime Minister and Minister of Digital Transformation Mykhailo Fedorov at an event on the occasion of signing the document in Kyiv on Tuesday.
According to him, the creation of the Ukrainian LLM will make it possible to collect a unique amount of data and work with defense-tech, process and generate certain solutions.
"Our scientists, the work of our scientists, our libraries will be there. Everything will be digitized and used to train this model. This model can be used so that various organizations, businesses, already trained to work with this model, can upgrade it, use it for their business purposes," Fedorov explained.
The development of the tool will last nine months, after which beta testing will take place during the year. The model will be created on one of the open-source solutions, without involving public funds. The first version is planned to be launched by December 2025.
Based on the language model, the state aims to create AI services for citizens and tools to optimize internal processes. The plans of the Ministry of Digital Transformation and WINWIN AI Center of Excellence include the launch of an AI assistant in Diia, AI tools for analyzing regulatory legal acts and working with European legislation, AI tools in Mriya for creating individual educational trajectories for schoolchildren, etc. In total, there are more than 10 products based on AI.
The Ukrainian-language LLM will provide more relevant and accurate answers for users than global models, as it will be additionally pre-trained on Ukrainian data. It is planned that thanks to a deeper understanding of the language specifics (dialects, terminology and context), it will be better oriented in many topics, in particular those related to Ukrainian history, culture and social experience. This will ensure not only linguistic, but also substantive accuracy of the answers.
According to Fedorov, a committee will be created to analyze what data the model is trained on, threats to it. In turn, Kyivstar will act as an operational executor of the process and form a project office for the development of the model: it will select a team, provide computing power for pre-training the Ukrainian model, etc. The company also fully covers the financial part of the LLM development, because the national LLM is developed at the expense of business without using the state budget.
