Inside the Belly of the Beast: How LLMs Work

About the session

In 2017, a landmark paper entitled “Attention is All You Need” turned the deep-learning world on its head and laid the foundation for today’s Large Language Models. In the paper, we learned about the transformer architecture and its unparalleled ability to understand, and even to generate, human language. Yet the inner workers of transformers are a mystery to most.

This session takes a deep dive into natural-language processing, starting with simple word embeddings and text classifiers, moving to Recurrent Neural Networks (RNNs), and ultimately landing on transformers.

You’ll come away understanding what a transformer is, how it works, and what happens as words flow through a transformer and are converted from static embeddings with low information density into rich, context-aware embeddings that capture the meaning of the words around them. You’ll also learn what context windows really are and why GPUs are so crucial to a transformer's operation.

It will give you a new appreciation for the brilliance of the transformer architecture. And it might make you a better engineer.