AI glossary

An AI system that can take actions toward a goal, such as searching, editing files or sending messages, not just answer questions.

The work of making AI systems pursue the goals and values their makers and users actually intend.

The part of a transformer model that decides which other words matter most when processing each word.

Asking the model to work through a problem step by step before giving its final answer.

Deciding everything an AI model sees for a task: instructions, examples, documents, memory and tool results, not just the prompt wording.

The maximum amount of text, measured in tokens, that an AI model can read and write in one go.

An AI mode that spends several minutes searching and reading many sources, then writes a detailed report with citations.

Markers like triple quotes or XML tags that separate your instructions from the text you want the AI to work on.

A type of AI that generates images or video by starting from random noise and gradually refining it into a picture.

Training a smaller AI model to imitate a larger one, so you get similar results for less cost.

Lists of numbers that represent the meaning of text, so a computer can find texts that mean similar things.

Tests that measure how well an AI model or prompt performs on a set of example tasks.

Showing the model two or three examples of the task done well before asking it to do a new one.

Training an existing AI model further on your own examples to change how it behaves.

A large, general-purpose AI model trained on broad data that many different apps and tasks are built on top of.

AI that creates new content, such as text, images, audio, code or video, rather than only sorting or predicting labels.

Making your content easy for AI answer engines like ChatGPT search, Perplexity and Google AI Overviews to find, understand and cite.

Making an AI base its answer on specific sources you provide or it retrieves, instead of only its memory.

Rules, checks and limits placed around an AI system to keep its behaviour safe and on task.

When an AI model states something false or made up as if it were true.

Running a trained AI model to get an answer. Every chat message you send triggers inference.

Training that teaches a raw language model to follow instructions and answer like an assistant.

Improving an AI result through a series of follow-up instructions, rather than expecting a perfect answer first time.

A prompt designed to trick an AI into ignoring its safety rules or usage policies.

The date after which an AI model has no built-in knowledge, because its training data stops there.

An AI model trained on huge amounts of text to predict and generate language, like the models behind ChatGPT and Claude.

Using an AI model to grade or compare AI outputs against criteria, instead of having people review every one.

A plain-text file at a website’s root that tells AI tools what the site is about and which pages matter most.

A setting that caps how long the AI’s reply can be, measured in tokens.

An open standard that lets AI apps connect to tools and data sources, like a universal plug for AI integrations.

Using an AI model to write, improve or critique prompts, rather than doing the task directly.

AI that can work with more than one type of input or output, such as text, images, audio and video.

A list of things you do not want in an AI-generated image, such as blur, extra fingers or text.

An AI model whose trained parameters are published, so anyone can download and run it on their own hardware.

The internal numbers a model learns during training. Model size, like "70B", counts them in billions.

Writing the first few words of the AI’s reply yourself, so it continues in the format you want.

The text you give an AI model to tell it what to do.

A feature that reuses the processed start of a prompt across requests, making repeated calls cheaper and faster.

Splitting a big task into a sequence of smaller prompts, where each step uses the output of the last.

Writing and testing instructions for an AI model so it reliably gives you the result you want.

An attack where hidden instructions in a document, email or web page trick an AI into doing something it should not.

A reusable prompt with blanks, like {{AUDIENCE}}, that you fill in each time you use it.

Storing an AI model’s numbers with less precision so it uses less memory and runs faster, with a small loss in quality.

A setup where the AI first searches your documents for relevant passages, then answers using them.

A pattern where an AI alternates between thinking about what to do, taking an action with a tool, and checking the result.

An AI model trained to think through a problem at length before answering, which helps on hard maths, code and analysis.

A training method where people rate AI answers and the model learns to give answers that get better ratings.

Telling the AI to act as a particular expert or character, such as an editor or a teacher.

Asking the same question several times, or several ways, and trusting the answer that comes up most often.

Search that finds results by meaning, not just by matching the exact words you typed.

A piece of text that tells the AI to stop writing as soon as it produces it.

Getting an AI to answer in a fixed format, like JSON or a table, that software or people can use directly.

When an AI model tells you what you want to hear, agreeing or praising instead of being accurate or honest.

Standing instructions given to an AI model before the conversation starts, which shape every reply.

A setting that controls how random or predictable an AI model’s word choices are.

A small chunk of text, often part of a word, that AI models read and write. Usage and prices are counted in tokens.

When an AI model asks to run a tool, like a search or a calculator, and uses the result in its answer.

A setting that limits the AI’s word choices to the most likely options that together add up to a set probability.

The huge collection of text, code and other material an AI model learned from.

The neural network design behind almost all modern language models, built around a mechanism called attention.

A prompting approach where the model explores several possible paths, judges them, and continues with the most promising.

A database built to store embeddings and quickly find the ones most similar to a query.

Building software by describing what you want to an AI and accepting the code it writes, often without reading it closely.

Asking an AI model to do a task with instructions only, without showing it any examples.