AI terms, explained

AI systems that don't just answer a question but plan and carry out a series of steps, using tools, to complete a goal.

AGI, or artificial general intelligence, is the hypothetical point at which an AI system can handle essentially any intellectual task a person can, rather than a narrow set.

Artificial intelligence is the broad field of building computer systems that do tasks we associate with human thinking, such as recognising images, translating text or answering questions.

Deep learning is machine learning that uses neural networks with many stacked layers, letting systems learn complex patterns in images, sound and language directly from raw data.

A foundation model is a large, general-purpose AI model trained on broad data that organisations adapt to many different tasks instead of building a new model for each one.

Generative AI describes systems that produce new content — text, images, audio, video or code — by predicting what plausibly comes next, based on patterns learned from large amounts of data.

A GPU is a chip originally designed for graphics that performs huge numbers of simple calculations at once, making it the workhorse for training and running AI models.

Inference is the stage where a trained AI model is actually used — taking your input and computing an answer — as opposed to the earlier, one-off training stage.

A large language model is an AI system trained on huge amounts of text that generates language by repeatedly predicting the most plausible next piece of a sentence.

Machine learning is the main way modern AI is built: instead of being programmed with rules, a system finds patterns in lots of examples and uses them to make predictions.

A neural network is a mathematical structure, loosely inspired by brain cells, made of layers of simple units whose connection strengths are tuned during training to recognise patterns.

An AI model whose trained parameters are published for anyone to download, so you can run it on your own computers instead of calling a company's service.

Parameters are the adjustable numbers inside a neural network that training tunes; they store what the model has learned, and their count is a rough measure of model size.

Training data is the collection of examples an AI system learns from; its size, quality and biases largely determine what the finished model can and cannot do well.