tech

Србија гради сопствени ВИ модел за српски језик

Председник Привредне коморе Србије (ПКС) Марко Чадеж и директор Канцеларије за ИТ и еУправу Михаило Јовановић потписали су данас Споразум о сарадњи у области дигиталне трансформације и развоја националног ЛЛМ модела, односно великог језичког ВИ модела за српски језик, који представља важан корак ка јачању дигиталног и ИТ суверенитета Србије.

Србија гради сопствени ВИ модел за српски језик

TL;DR

  • Serbia is developing its own national large language model (LLM) for the Serbian language.
  • This move aims to enhance Serbia's digital and AI sovereignty, reducing reliance on foreign systems.
  • The Serbian LLM will enable wider and more efficient AI application in public and private sectors.
  • Key benefits include keeping data and language within Serbia, preserving linguistic diversity (dialects and accents), and maintaining state control over AI development.
  • Serbia ranks high globally in digital infrastructure and AI readiness.
  • The LLM will support more effective public administration, reduce bureaucracy, and improve access to services.
  • It will also be used in education, healthcare, justice, and the economy, lowering costs and increasing business competitiveness.
  • The model is seen as vital for cultural identity and the future of the Serbian language.
  • Serbia plans to export its digital and e-government solutions, particularly to Africa, Central Asia, and South America.
  • Other institutions like libraries and media houses are invited to contribute to the LLM's development.
  • Serbia is acquiring a third supercomputer and a French system called Mistral to host and develop the Serbian LLM.
  • Access to supercomputer capacity and the Serbian LLM may be offered at preferential or free rates to domestic businesses.
  • Existing global AI models lack sufficient knowledge of the Serbian legal framework and specific language nuances in sensitive areas.
  • The development process is expected to take around six months for training, with initial results within a year.
  • The aim is to create an open model accessible to businesses and other AI tool providers, not just in Serbia but regionally.
  • The project leverages existing infrastructure like the Kragujevac supercomputer and data collection efforts.