Translation. Region: Russian Federal
Source: Peoples'Friendship University of Russia
An important disclaimer is at the bottom of this article.
Fernando León, a Master's student in Fundamental Informatics and Information Technology at RUDN University's Faculty of Physics, Mathematics, and Natural Sciences, took first place in the SMILES summer school on machine learning. The school was organized this summer by the Skolkovo Institute of Science and Technology in collaboration with the Harbin Institute of Technology (China).
Fernando's team, which also included representatives from Skoltech and the Higher School of Economics, prepared a project called "LLM Agents for Predicting Public Perceptions of Central Bank Actions."
The team developed a prototype synthetic focus group system based on large-scale language models (LLM). Instead of traditionally gathering real survey participants, the team created 10 virtual "avatars" with distinct socio-demographic characteristics—age, income level, occupation, and financial habits. These "avatars" responded to various announcements and initiatives from the Central Bank, such as changes to the key rate or support measures, allowing them to analyze how these news items were received by different segments of society.
"We worked on the project remotely for a week, choosing the topic from those proposed by the school organizers. The most challenging part for me was that it was my first time working with language models, but I was able to grasp the nuances fairly quickly. After completing the school, we further refined the project and improved our performance," said Fernando Leon, a master's student in the Faculty of Physics, Mathematics, and Natural Sciences at RUDN University.
A project by a RUDN University master's student team demonstrates how artificial intelligence can improve the speed, accuracy, and scalability of financial institutions' communications with society:
Virtual "avatars" provide feedback in minutes rather than weeks and require significantly fewer resources than traditional focus groups. They allow central banks to test and adapt their messages more quickly, increasing their clarity and reducing the risk of negative reactions. The project is easily scalable, and the creation of hundreds or thousands of "avatars" will allow for modeling the reactions of large populations.
For winning the competition, the team received 1 million tokens to use with GigaChat, Sber's artificial intelligence.
Please note: This information is raw content obtained directly from the source. It represents an accurate account of the source's assertions and does not necessarily reflect the position of MIL-OSI or its clients.