Artificial Intelligence

Talk about AI for long enough and you start to feel that everyone else was handed a glossary you missed. Inference, tokens, context windows, agents: the words arrive faster than the explanations, and most explanations are written either for engineers or for a sales meeting.

This section is for the person in between. Each explainer takes one term, says what it means in plain English, shows how it works with a diagram, and then does something the technical pages rarely bother to do: it tells you the word itself. Where it comes from, how its meaning drifted, what it collocates with, and how to say it in a sentence without sounding as if you swallowed a press release.

The AI vocabulary is also unusually slippery. “Model” can mean a file, a system, or a company’s product. “Training” and “inference” sound like synonyms and are opposites in almost every way that costs money. “Agent” is used for things that differ by an order of magnitude in what they can do. So the pieces are written to be read in an order: begin with the words that name the basic stages, then move to the ideas built on top of them.

Every page carries a date, a revision history and its sources, because this is a field where an explanation can go stale in a season. When something changes, the page changes, and the log says so.