I don’t Believe they Reason about Beliefs. Propositional Attitudes in Large Language Models

Alessandro Lenci

Università di Pisa

Le 11 décembre 2024 à 11 h 30

Réunion Zoom — Séminaire ILFC, voir lien plus bas


Large Language Models (LLMs) reveal outstanding linguistic abilities acquired through a simple, general-purpose word-prediction learning objective that allows them to extract huge amounts of knowledge encoded in distributional statistics. These so-called “emerging abilities” apparently include important aspects of pragmatic competence as well, such as for instance mastering indirect speech acts or interpreting figurative expressions. These phenomena are particularly important because they are usually considered to be the hallmark of human language understanding and creativity, and to presuppose sophisticated abilities to represent and reason about belief states. In this talk, I will present evidence from recent research investigating the performance of LLMs in pragmatic tasks. Results will be discussed within the general question of the scope and nature of language understanding in Generative AI.


Pour recevoir les annonces hebdomadaires par courriel, visitez http://rali.iro.umontreal.ca/rali/?q=fr/node/1631

Liste des autres séminaires