The episode explores the evolution of Large Language Models (LLMs), from the use of simple architectures to Transformer-based ones, highlighting the exponential increase in parameters and training data. It examines the challenges related to computational costs, data bias, and environmental sustainability, with a particular focus on applications across various sectors and the rise of multimodal models (MLLMs) capable of processing information from diverse sources (text, images, audio). Finally, it outlines future research directions centered on efficiency, reliability, and integration into complex ecosystems.