The CKL/CSE Seminar series continues on 18 November, providing a platform for researchers from Constructor Knowledge Labs (CKL), the School of Computer Science and Engineering (CSE), and the wider Constructor University community to engage in scholarly dialogue, exchange research insights, and explore opportunities for interdisciplinary collaboration across research and practice.
The upcoming seminar, “Same Model, Different Language,” will be delivered by Prof. Dr. Giancarlo Succi, Principal Investigator at Constructor Labs, Interim Dean of Computer Science at Constructor University, and Professor at the University of Bologna, Department of Computer Science and Engineering.
The seminar will present findings from four studies examining how the performance of large language models (LLMs) for code generation varies across programming languages. Drawing on data from three publicly available benchmarks — McEval, MultiPL-E, and ACB-Lite — the studies investigate language-specific performance differences, model selection across languages, changes in benchmark performance from 2024 to 2026, and similarities between programming languages based on model performance.
The findings reveal substantial differences in how coding models perform across programming languages, while also showing that a strong default model remains a reasonable choice for most languages. The studies further examine how model performance evolves over time, where language-specific model selection can improve accuracy, and how similarities in performance patterns can help identify relationships between programming languages.
The presentation will also address the methodological limitations of benchmark-based evaluations, including the difficulty of distinguishing language-specific challenges from the characteristics of benchmark problems. Together, the four studies emphasize that while benchmarks provide valuable guidance for model selection, the most meaningful evaluation of AI coding assistants ultimately depends on performance within a team’s own software development workload.
No registration is required. Online participants can join via the link provided below, while on-site attendees are welcome to come directly to the venue.
The seminar will be particularly relevant to researchers and PhD students working in computer science, software engineering, artificial intelligence, programming languages, and related interdisciplinary fields, offering an opportunity to exchange ideas, explore shared research interests, and foster new collaborations.
Date: 18 November 2026
Time: 13:00–14:00 CET (doors and online stream open at 12:45)
Format: On-site and online
Location: IRC Conference Room, Constructor University, Bremen
Online: https://constructor-org.zoom.us/j/97927329320?pwd=7kXzvbhm7awMbaiLv27xPKif53U1vr.1
Meeting ID: 979 2732 9320
Passcode: 591070
Agenda:
12:45–13:00 – Doors and stream open
13:00–13:05 – Welcome and Introduction
13:05–13:45 – Keynote Presentation by Prof. Dr. Giancarlo Succi
13:45–14:00 – Discussion and Q&A Session
