Far from being invented in late 2022, Large Language Models like ChatGPT find their roots in well established, multiple decades old literature in Machine Learning. In this talk, as such, Niccolo’ Gentile will first introduce the main algorithm behind said models, called Transformers, hence explaining what exactly happens from prompt inputting to sentence generation. Then, he will present one of his most recent publications, titled “Shaping Explanations: Semantic Reward Modeling with Encoder-Only Transformers for GRPO”, where said models, within an innovative training framework, are leveraged to address the famously challenging Italian Faculty of Medicine Admission test, with broader implications for pedagogical tasks
LISER had the pleasure to host the prestigious event which gathered over 80 participants
