AI glossary: 48 terms explained simply

Agent

An AI system that plans and uses tools in a loop to complete a multi-step task.

AGI

Artificial general intelligence: hypothetical AI with human-level ability across almost all tasks.

Algorithm

A step-by-step procedure for solving a problem.

Alignment

Making AI systems behave according to human intentions and values.

API

Application programming interface: a way for programs to talk to each other, such as an app calling an AI model.

Attention

The Transformer mechanism that lets each token weigh the importance of other tokens.

Backpropagation

The method of tracing errors backwards through a network to update its weights.

Bias (data)

Systematic unfairness in data or outputs that disadvantages certain groups.

Chatbot

A program that converses with people in natural language.

CNN

Convolutional neural network, a design especially good at images.

Context window

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

Dataset

A collection of examples used for training or testing.

Deep learning

Machine learning using neural networks with many layers.

Deepfake

Realistic fake media created with AI.

Diffusion model

A generative model that creates images or video by removing noise step by step.

Embedding

A list of numbers representing the meaning of text, images or other content.

Fine-tuning

Extra training of a pre-trained model on specific data.

Foundation model

A large general-purpose model adapted to many tasks.

GAN

Generative adversarial network: a generator and a discriminator trained against each other.

Generative AI

AI that creates new content such as text, images, audio, video or code.

GPU

Graphics processing unit, a chip that performs many calculations in parallel; widely used for AI.

Gradient descent

An optimisation method that adjusts weights step by step to reduce error.

Hallucination

A confident but false or invented AI output.

Inference

Using a trained model to produce an output.

LLM

Large language model: a model trained on vast text to understand and generate language.

Machine learning

AI that learns patterns from data rather than following hand-written rules.

MCP

Model Context Protocol: an open standard for connecting AI applications to tools and data.

Model

The trained system that maps inputs to outputs.

Multimodal

Able to process more than one type of data, such as text and images.

Neural network

A model made of layers of connected artificial neurons.

Open-weight model

A model whose trained weights are published so anyone can run it.

Overfitting

When a model memorises training data and performs poorly on new data.

Parameter

An internal number (such as a weight) learned during training.

Prompt

The instructions or question given to an AI model.

Prompt injection

An attack that hides instructions in content to manipulate an AI system.

RAG

Retrieval-augmented generation: retrieving relevant documents and adding them to the prompt.

Reasoning model

A model that works through problems step by step before answering.

Reinforcement learning

Learning by trial and error from rewards and penalties.

RLHF

Reinforcement learning from human feedback, used to make assistants more helpful and safe.

SDK

Software development kit: ready-made code libraries for using a service.

Supervised learning

Learning from labelled examples.

System prompt

Background instructions that define an assistant's role and rules.

Temperature

A setting that controls how random or creative a model's output is.

Token

A chunk of text (often part of a word) that a language model reads and writes.

Training

The process by which a model learns from data.

Transformer

The neural network architecture, based on attention, behind modern LLMs.

Unsupervised learning

Finding patterns in data without labels.

Vector database

A database that stores embeddings and finds similar items quickly.

Start the free AI course →