Events

Title: Mechanistically Interpreting Language Models: From Syntactic Code Completion to Arithmetic Reasoning

 Abstract:
Transformer-based language models (LMs) have shown remarkable progress in tackling complex tasks. However, their rapid advancements come with growing concerns about safety and reliability, stemming from our limited understanding of their inner workings. In this talk, I will share our recent efforts toward building a mechanistic understanding of LMs. First, I will present our study on why LMs struggle with a seemingly simple syntactic code completion task: balanced parentheses completion. We discovered that LMs fail not because of their lack of reliable mechanisms to accomplish the task, but that the faulty mechanisms inside them overshadow the sound ones. Next, I will share our findings about how LMs perform arithmetic on expressions such as “a + b - c”. Surprisingly, we found that LMs do not solve these equations in a human-like compositional manner. Instead, they transfer all information to the last-token position and rely on it to complete the calculation. Finally, I will conclude the talk with a brief discussion of our vision on mechanistically interpreting LMs.

Bio:
Ziyu Yao (https://ziyuyao.org/) is an Assistant Professor in the Department of Computer Science at George Mason University, where she co-leads the George Mason NLP group (https://nlp.cs.gmu.edu/). She works on LLM reasoning, planning, mechanistic interpretability, and human-LLM interaction, and has organized workshops on these topics at COLM, ACL, and NAACL. She recently gave a tutorial about Mechanistic Interpretability of LMs at ICML 2025. Her work has been funded by National Science Foundation, Virginia CCI, and Microsoft, among others. Prior to George Mason, she graduated with a Ph.D. degree in Computer Science and Engineering from the Ohio State University in 2021.

Para entrar en la reunión, pulse el siguiente enlace:
https://oficinavirtual.ugr.es/redes/SOR/SALVEUGR/accesosala.jsp?IDSALA=22999060
- Contraseña de la reunión: 180006

Title: Generative AI in Education: From Hype to Evidence

Abstract:
Generative AI entered the educational spotlight with intense, and often polarizing, reactions. For some, it signaled disruption and risk; for others, it promised transformative change long overdue. While initial debates were marked by speculation and uncertainty, the field has begun to shift. An expanding body of empirical research now offers more grounded insight into how generative AI is influencing learning and teaching. This talk will first examine how emerging evidence reveals the role of GenAI in enhancing educational practices. It will then address critical issues such as learner agency, overreliance, and the implications for assessment. Drawing on findings from a series of studies, the talk will argue for building a stronger evidence base and present a forward-looking research agenda to support responsible and informed innovation in educational contexts.

Bio:
Dragan Gašević is Distinguished Professor of Learning Analytics and Director of Research in the Department of Human Centred Computing of the Faculty of Information Technology and the Director of the Centre for Learning Analytics (https://www.monash.edu/it/colam) at Monash University. Dragan’s research interests center around data analytic, AI, and design methods that can advance understanding of self-regulated and collaborative learning.  He is a founder and served as the President (2015-2017) of the Society for Learning Analytics Research. He has also held several honorary appointments in Asia, Australia, Europe, and North America.  He is a recipient of the Life-time Member Award (2022) as the highest distinction of the Society for Learning Analytics Research (SoLAR) and a Distinguished Member (2022) of the Association for Computing Machinery (ACM). In 2019-2024, he was recognized as the national field leader in educational technology in The Australian’s Research Magazine that is published annually. He led the EU-funded SHEILA project that received the Best Research Project of the Year Award (2019) from the Association for Learning Technology. 

Para entrar en la reunión, pulse el siguiente enlace:
https://oficinavirtual.ugr.es/redes/SOR/SALVEUGR/accesosala.jsp?IDSALA=22999988
- Contraseña de la reunión: 428998