Large Language Models: What Works (Really Well) and What’s Next

by Roberto Navigli

Wednesday Jun 3, 2026, 4:00 PM → 5:30 PM Europe/Rome

Aula Conversi (Dip. di Fisica - Edificio G. Marconi)

Large Language Models (LLMs) have rapidly evolved from research prototypes into widely deployed systems that are transforming how we search, write, code, learn, and interact with knowledge. In this talk I will introduce the key factors behind the success behind LLMs and offer a pragmatic perspective on what these models actually do well today, highlighting the capabilities that have proven genuinely robust across applications, from reasoning over text and multilingual understanding to knowledge access and human-AI interaction. I will then examine the key limitations that remain unresolved, including factuality, reasoning reliability, efficiency, and grounding in real-world knowledge. Finally, I will discuss a few research directions that may shape the future of AI, including neuro-symbolic approaches, continual learning, multimodal intelligence, and more trustworthy and interpretable systems.