The EU Tries to Train a Model

The EU Tries to Train Its Own Model

TL;DR: An EU-backed project, OpenEuroLLM, aims to develop AI models for 32 languages, focusing on compliance with European digital laws, with a modest budget of €37.4 million over three years.

Key Points:

• Project Scope: Development of open-source AI models for 24 EU official languages and 8 candidate countries' languages.

• Funding: Budget of €37.4 million, significantly less than some large-scale AI projects in the U.S.

• Timeline: A three-year roadmap, with an intermediate model possibly available within a year.

• Participants: Includes companies like Aleph Alpha and Silo AI, supported by the European Commission and various research institutions.

Why It Matters:

The majority of LLMs been trained predominantly on English, Japanese, and Chinese data. This limits the opportunities for companies and public authorities in some countries to adopt AI as the models don’t understand the complexity of the language.

The EU with its varied countries and languages aims to address this. How they plan to accomplish this goal on $40 million dollars remains to be seen. I would have recommended they work with France’s Mistral and improve Mistral's AI to better support multi-lingual queries.