Inside LLMs: How They Work and Why They Scale

Deep learning has become the core engine of modern AI, enabling models to learn rich, non linear patterns directly from data. In recent years, this paradigm has scaled into a new generation of systems capable of generating text, solving problems, and interacting in natural language: Large Language Models (LLMs).

Participants will revisit, during this 3-hour online training, the core ingredients of deep learning, i.e. embeddings and non linear transformations, and see how these principles extend to models trained on massive text corpora. We then introduce the conceptual foundations behind modern LLM architectures, explaining at a high level how they process sequences, capture relationships between tokens, and scale with data and compute.

The web session highlights the key intuition behind the transformer framework without committing to a full technical deep dive. Attendees will gain a clear sense of why transformers replaced earlier approaches, how they enable efficient parallel processing, and discuss the practical implications of these models: what they are good at, where they struggle, and how their behavior is shaped by the underlying architecture.

The goal is to provide a compact, accessible, technically informed introduction to how modern language models work.

The objective of this 3 hour session is to build a solid, intuitive understanding of how modern Large Language Models (LLMs) are structured and why their architecture unlocks such powerful capabilities.