Seven chapters of pure on-ramp – what a model is, how it learns from examples, and why a chatbot is a prediction machine rather than a mind. No equations anywhere.

Vectors, matrices, derivatives, gradients, probability and entropy – each one introduced at the moment a machine-learning mechanism needs it, and landing on that mechanism running live.

What a model is, what it means to be wrong, how being-less-wrong becomes a procedure you can run – and why scoring well on the data you trained on proves nothing at all.

From a single artificial neuron to a trained network – what one unit computes, why stacking them buys expressiveness, how backpropagation assigns credit, and what depth actually learns.

Tokens, embeddings, attention as a data-dependent weighted average, the transformer block, and how a distribution over next tokens becomes text. The mechanism, not the product.