“Evaluating and Leveraging LLMs Across Diverse Educational Domains”, Dr. Marius Dumitran, University of Bucharest – Τμήμα Μηχανικών Η/Υ και Πληροφορικής

Την ερχόμενη Παρασκευή 9/5 θα έχουμε τη μεγάλη χαρά να φιλοξενήσουμε ως ομιλητή στα πλαίσια των εκδηλώσεων “Σεμινάριο CEID & Social Hour” και του ΔΠΜΣ ΥΔΑ τον Dr. Marius Dumitran, University of Bucharest.

Please note the following interesting and highly topical talk that will be presented in the context of the weekly event “CEID Seminar & Social Hour” organized by CEID, and the MS program DDCDM.

Τίτλος:  Evaluating and Leveraging LLMs Across Diverse Educational Domains

Ομιλητής: Dr. Marius Dumitran, University of Bucharest

Ημερομηνία-χώρος: Παρασκευή  9 Mαΐου,  3:15-5μμ, ΤΜΗΥΠ, αμφιθέατρο Γ

Abstract: This talk provides a multi-domain perspective on the capabilities and limitations of modern Large Language Models (LLMs) in educational assessment. Drawing on extensive evaluations across advanced algorithms, competitive programming, Romanian grammar, biology, and Baccalaureate exams, we present benchmarks for leading LLMs, demonstrating significant recent progress in accuracy and problem-solving. We explore novel applications, including using LLMs to automatically generate effective test cases for programming contests and to provide grammatical explanations, assessing the quality and reliability of these outputs through expert validation. Key findings include the identification of specific LLM strengths (e.g., theoretical problems) and weaknesses (e.g., visual tasks, explanation fidelity), the importance of curated datasets for low-resource languages, and the successful development of experimental platforms and human-AI collaborative methods. While LLMs offer powerful tools for enhancing assessment and feedback, our research emphasizes the continued need for human oversight and highlights directions for future work in creating robust, reliable, and pedagogically sound AI-assisted educational tools.

About the speaker: Dr. Marius Dumitran is a computer scientist and lecturer at the University of Bucharest, whose industry experience spans software‑engineering roles at Google, Twitter, Meta and Palantir. A Forbes “30 Under 30” honoree in Education, he has presided over Romania’s National Olympiad in Informatics—coaching seven teams to ACM‑ICPC World Finals—and continues to mentor high‑performing competitive‐programming students. His research focuses on data structures, algorithms and on leveraging artificial intelligence in education; his recent work has been accepted or is under review at venues including ITS, ACL and BEA 2025.