Looking beyond AI benchmarks

Large language models can write essays, answer questions, and solve increasingly complex tasks. But how much do these systems actually understand – and how reliable are they once they leave controlled test environments?

These are the questions that drive the research of Tamara Stojanovska, who recently joined the Chair of Artificial Intelligence and Society (AISOC) as a PhD researcher under the supervision of Prof. Bilal Zafar.

Her work focuses on uncovering the deeper capabilities of large language models (LLMs). Rather than relying solely on standard AI benchmarks, Tamara investigates how such systems perform in more complex, real-world situations – where understanding, reasoning, and generalization become much harder to measure.

A particular focus of her research lies on the reliability and trustworthiness of AI agents. As AI systems increasingly interact with users, tools, and online environments, questions of robustness and safety become more important. Tamara’s goal is to help develop methods that allow these systems to be deployed in ways that are dependable and aligned with human expectations.

Her research fits closely with the broader mission of AISOC and the Research Center Trustworthy Data Science and Security: building AI systems that are not only capable, but also understandable, reliable, and safe to use in practice.

Outside of research, Tamara enjoys playing volleyball, cooking, and inline skating.

Welcome to RC Trust, Tamara!