This summer, the Federal Polytechnic Schools of Lausanne and Zurich will provide a large model of language (LLM) developed on public infrastructure. Trained on the “Alps” supercaluler, this new model of artificial intelligence is distinguished by its ethical and open approach.
Swiss scientists unveiled on Wednesday In Geneva a new LLM model, fully open and publicly developed. The technology, which was designed using the “Alps” super-computer of the Swiss Center for Scientific Calculation (CSCS) in Lugano, focuses on transparency, multilingual performance and wide accessibility.
Currently in the final test phase, it will be downloadable for free and freely later this summer. The model marks a turning point for open source artificial intelligence (AI) and multilingual mastery, according to its developers.
Revolution in the world race at AI
Since the arrival of the first LLMS accessible to the public, models supplying Chatgpt, Gemini or Deepseek have turned many sectors upside down. Their specialty: respond in natural language, write texts or translate, thanks to training on billions of words.
The new Swiss model, co -designed by scientists from the two EPFs and other Swiss universities in collaboration with the CSCS, marks a turning point in the global competition around AI. For experts, this is a key step towards a global collaborative ecosystem around transparent models, facing the trade systems dominated by the United States and China.
Reliability and responsibility at the center
The source code and training data will be transparent and reproducible, in order to encourage adoption in the scientific, government, educational and private fields. This approach aims to promote both innovation and responsibility.
The development of the LLM takes into account the Swiss data protection laws, Swiss copyright and transparency obligations within the framework of the EU EU regulations.
“Fully open models allow high confidence applications and are essential to advance research on the risks and opportunities of AI. Transparent processes also facilitate regulatory compliance,” explains Imanol Schlag, researcher at the IA center of the EPFZ, who pilot this project alongside Antoine Bosselut and Martin Jaggi at the EPFL.
A model that speaks more than 1000 languages
One of the characteristics of this LLM is that it masters more than 1000 languages. “From the start, we wanted the models to be systematically multilingual,” said Antoine Bosselut, quoted Wednesday in a press release from the two EPFs.
Unlike American technologies that focus on English, the basic model training has been carried out on a large corpus of texts in more than 1,500 languages - around 60% in English and 40% in other languages - as well as code and mathematics data. Thanks to this linguistic and cultural representation, the model allows you to be applied all over the world.
KSO with ATS
