How LLMs work, part 1: from the perceptron to reasoning models

Today's language models write code, solve olympiad problems and reason step by step. Every one of those abilities rests on ideas that are decades old. This first part of our seven-part series walks the timeline from a single artificial neuron in 1957 to the reasoning models of 2025, so the rest of the series makes sense: why models look the way they do, and which problem each idea solved.