The webinar introduces the fundamental principles of Large Language Models (LLMs), showing that their core mechanisms are not as complex as they may seem, while also highlighting how these mechanisms can give rise to various types of failures.
The webinar “Debunking LLMs Exhibiting the Known Sources of Possible Weaknesses: Hallucination, Fairness, Reliability“, presented by Prof. Frédéric Precioso (Université Côte d’Azur), examines recent advanced techniques, such as self-distillation and Reinforcement Learning from Verifiable Reward (RLVR), noting that while these methods do not fully resolve known limitations, they open new application areas and enable more effective use of LLMs. The session also offers insights into current evaluation approaches and the ongoing challenges related to model reliability. Finally, it considers the ethical implications of these technical choices, particularly regarding AI bias and fairness.
“Debunking LLMs Exhibiting the Known Sources of Possible Weaknesses: Hallucination, Fairness, Reliability” is part of the VOILA! Seminars organised by EFELIA Côte d’Azur – French School of Artificial Intelligence. These seminars aim to explore the frontiers of AI in an inclusive and open manner, welcoming everyone. The goal is to provide insights and answers to major societal and academic questions on topics such as AI & Environment, AI & Work, AI & Education, AI & Media, AI & Law, AI & Creativity, AI & Health, and much more.
Frédéric Precioso is a Professor of Computer Science at Université Côte d’Azur. He has made methodological contributions to areas of artificial intelligence, machine learning, and deep learning, such as explainability, hybridization of symbolic and non-symbolic AI, knowledge injection, low-data domains of expertise (medicine, social sciences).
Professor Precioso was appointed to the French National Research Agency (ANR) where, from 2018 to 2023, he was in charge of programs associated with the National AI Plan within the Directorate of Major State Investment Projects and the Department of Digital Sciences and Mathematics. He was responsible for all ANR funding instruments related to the French artificial intelligence strategy.
