AI glossary

Definitions written for someone who has to make a decision, not pass an exam. Each entry says what the term means, when it matters and what it does not solve.

All A C E F H I L M P Q R S T V

A

  • Agent

    A model given tools and a goal, allowed to decide its own next steps in a loop rather than answering once.

C

  • Chain of thought CoT

    Prompting a model to write out its intermediate steps before the final answer.

  • Context window

    The maximum amount of text, measured in tokens, that a model can take into account at once - your prompt and its answer together.

E

  • Embeddings

    Numeric vectors representing text, where similar meanings land close together - the basis of semantic search.

F

  • Few-shot prompting

    Showing the model two or three worked examples of the task inside the prompt, instead of only describing it.

  • Fine-tuning

    Further training of an existing model on your own examples, to change its behaviour rather than its knowledge.

H

  • Hallucination

    A confident, fluent answer that is simply wrong - an invented citation, function, statistic or fact.

I

  • Inference

    Running a trained model to produce output - what happens every time you send a prompt.

L

  • Large language model LLM

    A model trained on large amounts of text to predict the next token, which in practice lets it write, summarise, translate and reason over text.

M

  • Multimodal

    A model that accepts or produces more than one kind of input - text plus images, audio or video.

P

  • Prompt injection

    An attack where instructions hidden in content the model reads get executed as if you had written them.

Q

  • Quantisation

    Storing a model with lower-precision numbers so it fits in less memory and runs on ordinary hardware.

R

  • Reasoning model

    A model trained to spend extra computation working through a problem before answering.

  • Retrieval-augmented generation RAG

    A pattern where relevant documents are fetched first and pasted into the prompt, so the model answers from your data rather than its training.

S

  • System prompt

    Instructions given to the model outside the conversation, which apply to every turn rather than one message.

T

  • Temperature

    A sampling setting that controls how much randomness goes into choosing each next token.

  • Token

    The unit a model reads and writes - roughly a word fragment. Pricing, context limits and speed are all measured in tokens.

V

  • Vector database

    A store built for finding the nearest vectors to a query vector quickly, across millions of items.