Introduction

By the end of this lesson, you will understand how LLMs work under the hood, starting with how text is broken into tokens and converted into lists of numbers (vectors) through tokenisation and embedding, then how the transformer architecture processes those vectors to find connections between words and generate new text.

You will also explore how reinforcement learning from human feedback shapes a raw model into a helpful assistant.

You will learn why every conversation starts from scratch due to the stateless nature of LLMs and the limits of the context window. And you will see how grounding connects AI outputs to real-world facts.

Finally, you will look at how reasoning models think before they answer, and examine the deeper question of whether LLMs actually understand anything at all.