Lecture series: How Large Language Models (LLMs) Work

The landscape of Natural Language Processing (NLP) has been revolutionized by neural networks—particularly by large language models (LLMs) built on Transformer architecture. This course offers a comprehensive exploration of the theory and technology behind cutting-edge models like ChatGPT. The lecture series is led by Prof. John Abela from the University of Malta, an expert in NLP and machine learning.

We begin with the foundations of NLP, tracing its evolution from symbolic approaches to neural networks. Key concepts include vector space models, word embeddings, and the shift from static to contextual representations using models like Word2Vec, GloVe, FastText, and BERT.

The course then delves into the architecture of Transformer models, with a detailed look at GPT-3. Topics include self-attention, positional encoding, training dynamics, and the role of scale in unlocking capabilities like few-shot learning. We also explore emerging architectures such as Mixture of Experts (MoE).

Designed to bridge theory and practice, this course equips participants with the knowledge needed to understand, build, and interpret modern LLMs—grounded in both machine learning principles and mathematical foundations.

The week kicks off with a welcome drinks reception and includes a guided city tour of Groningen, offering participants the chance to connect, explore, and enjoy the vibrant academic and cultural atmosphere of the city. More information about the lecture series can be downloaded in the PDF below.