Artificial Intelligence

How an AI agent works, from the language model inside it to what it means for us

Everybody talks about AI agents: programs that get a goal instead of a recipe, plan their own steps, use tools, and keep going until the job is done. You will watch one at work, try one yourself, and then open it up, layer by layer.

We start at the top, with the agent you have just seen in action, and work our way down: to the language model inside it, to how that model represents text and meaning, to how it was trained, and finally to the simple idea of learning from data. Then we climb back up and ask what all this means for your field and for the rest of us.

Every session starts with an experiment. You do something first and understand it afterwards, often with pen, paper and a table before any computer comes in. All experiments can be repeated at home. Where there is code to write, you don’t type it yourself: you tell an AI agent what you need and check what it builds.

What you find here

The experiments come first, with step-by-step instructions as a web page and as a PDF to download, and the concepts behind them follow, each with its slides. Everything you need for the course is on this site.