Glossary

The bedrock ideas — what "AI" actually refers to, and how machines learn from data instead of rules.

How the chat assistants work under the hood — tokens, attention, context, and the transformer.

How raw models are built and then shaped — from pretraining on the open internet to preference tuning and cheap specialisation.

The practical dials and habits for getting reliable results out of a chat model.

Models that think before answering, and systems that act — using tools, writing code, and looping toward a goal.

Feeding models the right facts at the right moment — the machinery behind search-grounded answers and "chat with your docs".

Beyond text — models that see, hear, read images, and speak.

How prompts become pictures and clips — diffusion, the dials that steer it, and the pieces of a local pipeline.

What can go wrong, and the work of making models behave — honestly labelled, including the parts that are unsolved.

The practical side of using models — hardware, speed, cost, APIs, and the difference between open weights and open source.

How models are measured — and why the leaderboards deserve a raised eyebrow.

The organisations behind the models and tools — who they are, what they ship, and the honest caveats about hype, governance and lock-in.

The named assistants, image and video generators, and local runners people actually use — Claude, ChatGPT, Gemini, Stable Diffusion, Ollama and the rest.

The connective tissue between your code and the models — HTTP, JSON, keys, and webhooks, the plumbing under every "talk to an AI" feature.

Wiring AI into pipelines that run without you.

The languages, editors, and version control that AI-era developers actually work in every day.

The operating layer under almost every AI server and self-hosted tool — quietly running things whether you notice it or not.

The storage layer under every AI application — where the data (and increasingly the embeddings) actually live.

The arithmetic underneath the magic — vectors, matrices, and the handful of operations every model reduces to, explained without requiring you to do any of it yourself.

The silicon AI actually runs on — the chips, memory, and wiring that turn algorithms into watts and dollars.

The unsettled questions around AI — who owns the training data, what the law now requires, and where the technology helps or harms real people. Contested ground, presented fairly.

The vocabulary of the discourse — the memes, insults and shorthand people use to argue about AI online. Wry where it earns it, fair where it counts.