"She graduated from Fazekas High School in Budapest last year, but on the eve of her oral exams, she was still working with her colleagues on the scientific article that ultimately earned them an invitation to the NeurIPS 2025 conference in California. We caught up with 19-year-old Christ Miranda Anna and her mentor, Adrián Csiszárik, for a short chat," writes journalist István Balla in HVG. The article discusses the huge success of the Rényi Institute's AI group being accepted to the event in San Diego this year, and in one of the featured categories at that. "The professional jury selected the Hungarian researchers' paper from among the submitted articles to be included in the top 3% – the so-called Spotlight program. This means that Christ Miranda Anna, Csiszárik Adrián, Becsó Gergely, and Varga Dániel were able to present their results last week to Meta, Google, and other leading IT companies, as well as the scientific elite in the field."
The journalist points out that Anna Christ Miranda was probably the youngest researcher invited to this serious conference, who is currently a student at the University of California (UC Berkeley). Miranda has been interested in artificial intelligence since she was in tenth grade, and it was through a school assignment related to this topic that she met ELTE lecturer and Rényi Institute colleague Adrián Csiszárik. Miranda was accepted into the Rényi Institute while still in high school, before beginning her studies at the American university, and thus became a member of the National Artificial Intelligence Laboratory group operating at the institute.
HVG magazine reports on their joint work and the creation of their highly successful study. Adrián Csiszárik tells the journalist: “AI models are trained using some kind of method, but while they are being trained, and then when we use them, we have no control over what is happening inside them. You could say it’s a black box. Simply put, large language models work by learning to predict the next word. But while they are learning this, somehow the connections of the world are stored in this black box. This is what we hardly understand at all: what is really happening in there? How is the world represented in this black box?" A separate field of science has recently emerged to research the interpretability of AI, and the article by Hungarian researchers answers a sub-question in this field, as we can read in HVG, where Anna Christ Miranda summarizes the essence of their study: "There are large language models, such as ChatGPT, whose internal world model learns connections from the world through so-called relations. We examined how these relations are structured in large language models..."
You can read the full article HERE
