University of Turku and SiloGen launch consortium to build the world’s largest open LLM

The SiloGen and TurkuNLP consortium aims to establish the world’s largest open-source language model encompassing all official European languages. This initiative prioritizes digital sovereignty by ensuring that underlying models are trained on data accurately representing European citizens and organizations, while strictly adhering to regional regulations. By creating transparent and reproducible foundation models, the project seeks to guarantee accountability and prevent reliance on external, non-compliant technologies. Central to this effort is the unique combination of world-class academic expertise, curated multilingual datasets, and access to Europe’s most powerful supercomputing infrastructure. This alignment allows researchers to efficiently train large-scale models that prioritize multilinguality and equity across all languages. The project demonstrates how high-performance computing resources can be leveraged not just for scale, but for creating representative and culturally appropriate artificial intelligence systems. This initiative is highly relevant to the open data community as it validates the critical role of high-quality, representative public data in building trustworthy AI. It highlights a shift toward decentralized, sovereign technology development where open access to foundational models empowers downstream applications. By making these models freely available, the consortium democratizes access to advanced AI, enabling diverse organizations to innovate securely while maintaining control over their specific data and use cases.

Source: utu.fi
Published on 2023-10-20

Open Data

Related news