A Large Language Model (LLM) summer school was held at the American University of Armenia, with 70 students participating.
Jiao Sun, an AI researcher from Google’s Gemini AI team, arrived in Armenia to give lectures at the summer school. In a conversation with iTel.am, she mentioned that before coming, she had been very interested in Armenia and had studied Armenian culture.
“I’ve known Aram Galstyan, the co-organizer of the summer school, since my student years at the University of California. He knows my areas of interest and suggested that I join this project.
[[gallery1]]
I work with students, particularly in developing modern technologies for AI instructions, as well as evaluating the resulting models. My advice to them is not to fall behind the rapidly developing pace of technologies and to focus on improving the architecture of already created models, to think about what innovation they can add to current trends,” notes Jiao.
With the support of Picsart company and tech entrepreneur Vahe Kuzoyan, leading scientists have been invited to Armenia from the University of Southern California (USC), Massachusetts Institute of Technology (MIT), Google Research, University College London (UCL), University of Copenhagen (UCPH), Amazon, and Meta companies.
[[gallery2]]
The co-organizer of the summer school and head of the AI research team at ServiceNow company in Canada, Perouz Taslakian, says that initially they planned to involve 40 students, but the number of applications exceeded expectations.
“I taught at the American University for about 3 years and was the program chair of the BS in Computational Sciences. I saw the potential of young people here. After moving to Canada, since 2015, I have been organizing various conferences and workshops almost every year, inviting global experts and discussing open problems in the field. This year, Aram Galstyan and I decided to implement an initiative in the direction of Large Language Models, as we are both currently working in the AI field.
[[gallery3]]
The goal was to bring new knowledge to Armenia using our network and provide information about modern technological developments and research. It was very important for us to have practical classes, as most of the lecturers work in large companies,” says Perouz.
[[gallery4]]
For practical classes, it is necessary to have appropriate computing power (GPUs), which, according to Perouz, is very expensive. The European company Nebius AI has provided the necessary computational infrastructure for conducting the school. Each student was given access to one H100 GPU throughout the school.
Anastasia Zemvskova is from the Nebius AI team and is also in Armenia with her Armenian colleague to support the course of the classes. In her assessment, the level of the material being taught is complex, and the attendees are not ordinary students, but future or already active scientists.
[[gallery5]]
“We have built an AI-focused cloud platform tailored for machine learning engineers so that they can develop everything related to AI and ML, from experimentation to training. We were testing different business models, and as soon as we tried machine learning and GPU as a service model, it turned out that there is a huge shortage and demand in the market. We have our own data center in Finland, and we're not just reselling GPUs, but developing new systems around it.
[[gallery6]]
When I met Hrant Khachatrian from YerevaNN, he presented his research, which was very interesting because our machine learning team is also conducting several studies that complement each other. This is how we started collaborating,” adds Anastasia.
The YerevaNN foundation supported the organizational work of the summer school. The American University of Armenia hosted the school.
All the details of the summer school are available here.
