Using AI well

This module is optional. It is a primer for anyone who has come to the course without much experience of AI: what it is, the basics of using it, and what is possible. It needs no technical background and takes about 45 minutes. If you already use an AI assistant most days, start with practical AI.

This module covers the craft of building with models: the instruction files, skills and plugins that shape an agent, models that take more than text, connecting a model to tools and systems, answering from your own documents, and keeping the result safe and measured. It names products and file formats throughout, and those date quickly. It is where the course starts, and it needs no knowledge of how a model works. Its first part lists the seven things about a model’s behaviour that the rest leans on. Its part 6 ends with a design review checklist and the module’s own two reference sections. The whole module is about three hours of reading.

This module takes the practice from one person to a team and an organisation. Part 1 goes along the whole software lifecycle, from discovery to operations, and shows that most of the gain is outside coding. Part 2 is about redesigning how a team works so that the AI’s speed reaches delivery, and how to measure whether it has. Part 3 covers the decisions above the team, vendors, risk, governance and the investment case, and how to explain all of it to any audience. It ends with the module’s two reference sections. The whole module is about two hours of reading.

This module is what happens behind the scenes. Part 1 follows a prompt through the machine. Part 2 explains how a model comes by its abilities and its habits. Part 3 covers what it costs to run and how it fails, and ends with the module’s two reference sections. None of it is needed to use the tools well. Every technical idea is included because it explains a behaviour you have seen or a decision you will make. The whole module is about two hours of reading.

This module is hands-on, and it dates faster than the others. It covers the hardware, where models are published and how to read their names, the software that runs them, agents and the harnesses that drive them, and building on a local model. It names products throughout and says when each table was written. It explains what the tools do and leaves the commands to their own documentation. It leans on part 3 of the language models module, and can be read without it. The whole module is about two and a half hours of reading.

This module covers models that generate images, video, music and 3D objects. Part 1 explains the machine they share, and the second design that competes with it. Parts 2 to 5 take each medium in turn. Part 6 covers how models are adapted with fine-tunes, LoRAs, control inputs and style transfer, where to run them, and the questions of rights and provenance. Part 7 covers how to prompt them. You can read this module without the first. The whole module is about four hours of reading.

This module will grow. Each part stands alone, and each names products and figures that date quickly. It covers kinds of model that the other modules do not: small models such as the Smol, Phi and Needle families, world models such as NVIDIA’s Cosmos, DeepMind’s Genie and World Labs’ work, language models built on diffusion, and models such as Jev that return a decision and no text. Part 4 describes a product announced days before it was written, from its maker’s own figures, and says so. The whole module is about two hours of reading.