Glossary

Software that performs tasks we normally think of as requiring human judgement.

A way of building software that learns patterns from examples instead of being given rules.

The trained file that holds everything an AI system learned.

The process of showing a model huge amounts of data so it can learn patterns.

A defined set of steps for solving a problem.

A model structure made of many simple layers, loosely inspired by the brain.

The moment a trained model is actually used to produce an answer.

A specialised chip that performs many calculations at the same time.

A model trained on enormous amounts of text to predict what words come next.

A small chunk of text that a model reads and writes, usually part of a word.

The instruction and information you give a model.

The maximum amount of text a model can consider at one time.

A setting that controls how predictable or varied a model's output is.

When a model states something false with complete confidence.

Additional training that adapts a general model to a specific job or style.

Hidden instructions that set a model's role and rules before you type anything.

An AI system that can take steps and use tools to complete a goal, not just answer.

Letting a model call outside software to get facts or take action.

The mechanism that lets a model request a specific piece of software with specific inputs.

A model working through a problem in steps before giving an answer.

A fixed sequence of steps an AI system follows.

How much an AI system is allowed to do without asking a person.

Deliberately requiring a person to review or approve an AI action.

A collection of examples gathered for a model to learn from or be tested on.

Structured data lives in neat rows and columns; unstructured data does not.

A list of numbers that captures the meaning of a piece of text or an image.

A database built to store embeddings and find the closest matches fast.

Fetching relevant documents first, then asking the model to answer using them.

An agreed way for two pieces of software to talk to each other.

Information about data rather than the data itself.

The frontend is what the user sees; the backend is the part that runs on servers.

Making a new version of software live for real users.

The delay between a request and its response.

Systematically testing whether an AI system's output is actually good.

Rules and checks that constrain what an AI system may say or do.

What you pay for each chunk of text a model reads and writes.

A model systematically favouring or disadvantaging certain people or outcomes.

Any information that can be traced back to a specific person.

How long a service keeps what you send it, and what it does with it.

Hiding instructions inside content so a model obeys the attacker instead of you.

How well an AI system's behaviour matches what people actually intend.

How much is publicly known about how a model was built and what it was trained on.