Large Language Models can seem magics, but in fact they are built on layers of math, research, and innovations over the years. In this session, we'll cover the math behind them and see how you could build one if you had unlimited time, money, and training data.

We'll start with a simple bigram prediction model and work our way up. We'll cover neural network such as backpropagation and gradient descent. By the end of this talk, you'll see how you could build a very tiny GPT.