Director of ILSP Mr. Vasilis Katsouros on Action 24
05-06-2025
The Director of the Institute for Language and Speech Processing (ILSP) of the Athena Research Center, Mr. Vasilis Katsouros, appeared on the Action 24 channel and the show "ACTION HI-TECH", where, in two consecutive episodes, he presented the journey and vision behind the development of Greek Large Language Models (LLMs) at Athena. He discussed the first Greek LLM, Meltemi, introduced by the Institute about a year ago, as well as a more recent model, Llama-Krikri.
You can watch the first part of the program here:
During the discussion, he explained what Large Language Models are, how they are created, and what they require: vast amounts of high-quality data, specialized algorithms, and significant computational infrastructure. He emphasized that creating an LLM is not an easy task and requires continuous training to improve. He pointed out the challenges and issues involved, how they can be addressed, and what developments are expected in the near future.
He then referred to the advantages of the Greek LLMs developed by Athena, which are designed to be energy efficient and suitable for Greek industry and research. Mr. Katsouros also talked about the challenges of this effort, such as human resources and access to computational resources, stressing the importance of interdisciplinary collaboration for the success of such initiatives.
The existence of Greek LLMs is "particularly important for the digital survival of the Greek language," he notably stated, highlighting the need for the Greek language to be on equal footing in the new technological landscape.
Finally, the episodes touched upon the next steps through initiatives in which the Athena Research Center is involved, such as the Faros -Greek Artificial Intelligence Factory and the Alliance for Language Technologies European Digital Infrastructure Consortium, aiming to ensure that "the Greek language is represented in open multilingual LLMs for Europe."
You can watch the second part of the program here: