MISANU

PROGRAM

Predavanja možete pratiti i online putem MITEAM stranice Seminara iz veštačke inteligencije:
https://miteam.mi.sanu.ac.rs/asset/DPP9i2jhvYzp8dmRe

Plan rada Seminara iz veštačke inteligencije za NOVEMBAR 2026.

Sreda, 04.11.2026. u 19:00, Online
Danka Jokić, Milica Andrić, Nebojša Damnjanović, Toma Knežević, University of Belgrade, Belgrade, Serbia; Faculty of Philology, University of Belgrade, Belgrade, Serbia
BUILDING A SERBIAN RIDDLE BENCHMARK FOR MULTICULTURAL EVALUATION OF LARGE LANGUAGE MODELS
Large Language Models (LLMs) perform impressively across a wide range of NLP tasks. However, their grasp of figurative language, culturally grounded knowledge and commonsense reasoning remains underexplored, especially in low-resource languages. Riddles are a demanding testbed for these abilities. Solving them depends on indirect clues, ambiguity and shared cultural knowledge rather than on linguistic fluency alone.
In this talk, we present the Serbian contribution to a multicultural riddle benchmark initiated by Cohere Labs. The project brings together more than 40 language teams, in which native speakers act as creators rather than translators. Our contribution covers three parts:

Main Collection: 100 Serbian riddles across ten thematic categories. Eight categories are shared across languages, and we chose two to reflect Serbian culture.
Annotation: native-speaker estimates of each riddle's difficulty.
Separate Pilot evaluation: a separate held-out set of 30 riddles. Gemini 3.1 fully or partially solved 26 of them, and GPT-4o solved 19.

We also discuss a recurring failure mode. Models often translate a Serbian riddle into English before solving it, and in this process lose culturally grounded clues.

Sreda, 18.11.2026. u 19:00, Online
Vladan Devedžić, Serbian Academy of Sciences and Arts
QUO VADIS, EDUCATION, IN THE AI ERA?
This presentation addresses the transformative, yet unpredictable impact of Generative AI on higher education. While GenAI offers unprecedented capabilities - spanning automated assignment grading, gamified lesson creation, virtual teaching, and administrative support - the rapid evolution of AI technologies raises fundamental questions about the future trajectory of educational frameworks.
Current trends indicate a landscape in flux. Although the initial hype around AI is gradually settling and giving way to AI fatigue, advancements in large language models (LLMs) continue to accelerate. Capable of performing complex tasks faster than humans, AI systems and autonomous agents are reshaping higher-education course design and practices. Consequently, traditional pedagogical paradigms are forced to adapt. Higher education is shifting from knowledge-acquisition to skills-based instruction, where practical execution takes precedence over foundational knowledge, and personalized AI-assisted schooling challenges standard classroom structures.
Despite these opportunities, profound uncertainty looms over what institutions should teach, how curricula should be structured, and how quickly these changes must occur. Educators face non-trivial educational, ethical, legal, financial, and societal challenges. Questions persist regarding how to update core required courses, integrate elective tools like retrieval-augmented generation (RAG) and prompt engineering, and reimagine classic subjects like programming in an era dominated by automated coding. Illustrated through multiple examples from the presenter's teaching practice, the presentation discusses a growing need for continuous adaptation to an ever-evolving technological ecosystem powered by rapid developments in AI.

Sreda, 25.11.2026. u 19:00, Online
Branislav Kisačanin, Institute for Artificial Intelligence Research and Development of Serbia
UPDATE ON AI FOR REASONING IN MATH AND SCIENCE
We will discuss the latest developments in AI for mathematics and other applications of reasoning. Bring your hardest questions, we will reason through them together.

Ovaj onlajn seminar nastao je kao nastavak sastanka "Serbian AI Meeting" i zamišljen je da na njemu istraživači iz Srbije i iz dijaspore, kao i istraživači sa univerzteta, naučnih instituta i iz prakse predstavljaju naučne teme i rezultate iz oblasti veštačke inteligencije.
Link za svako pojedinačno predavanje biće dostavljen dan pre održavanja predavanja.

Andreja Tepavčević
Rukovodilac seminara

Biljana Stojanović
Sekretar seminara