Every AI term you'll actually run into, defined in one sentence first, then explained properly. No prior knowledge assumed.
Software that performs tasks we'd normally say require human judgement — recognising, predicting, generating, or deciding.
A set of step-by-step instructions a computer follows. In AI, it usually refers to the specific method used to train or run a model.
A way for one piece of software to talk to another. Most AI tools offer an API so developers can build their own apps on top of a model, rather than using the chat interface.
When a model's output unfairly favours or disadvantages certain groups, usually because its training data reflected the same imbalance.
A program designed to hold a conversation with a person, usually through text. Modern chatbots like ChatGPT are powered by large language models.
The amount of text a model can "see" and consider at once when generating a response — including your prompt and the conversation history.
A technique within machine learning that uses layered structures (neural networks) loosely inspired by the brain. It powers most modern AI, including language models and image generators.
A way of converting words, images, or other data into a list of numbers that captures their meaning, so a computer can compare how similar two things are.
Taking an already-trained model and training it further on a smaller, specific dataset so it performs better on a particular task.
AI systems that create new content — text, images, audio, or code — rather than just analysing or classifying existing data.
A type of computer chip originally built for rendering graphics, which turns out to be extremely good at the math behind training AI models. Most large models are trained on thousands of GPUs.
When an AI model generates information that sounds plausible but is factually incorrect or entirely made up — including fake citations, statistics, or quotes.
A model trained on enormous amounts of text, used to predict, write, and hold conversations in natural language. ChatGPT, Claude, and Gemini are all built on LLMs.
The main approach behind modern AI — systems that improve by studying examples, instead of following rules a person wrote by hand.
The end result of training — a system that has learned patterns from data and can now make predictions or generate output based on them.
A structure of interconnected nodes ("neurons") loosely modelled on the brain, used to process data in deep learning systems.
The branch of AI focused on understanding and generating human language — the foundation behind chatbots, translation, and text analysis.
When a model memorises its training examples too closely, performing well on data it's seen but poorly on new data it hasn't.
The internal numeric values a model adjusts during training. A model with "more parameters" is generally larger and more capable, though not always better for every task.
The instruction or question you give an AI tool. "Prompt engineering" refers to the skill of writing prompts that reliably get better results.
A technique where a model looks up relevant documents or data before answering, so its response is grounded in real information instead of relying purely on what it memorised during training.
A training method where a system learns through trial and error, receiving a reward signal for outcomes that are considered good.
Training a model using labelled examples — data that's been tagged with the correct answer, so the model learns to match new inputs to the right output.
A small chunk of text — often a word or part of a word — that a language model processes as its basic unit. Models are typically priced and limited by token count, not word count.
The process of adjusting a model's parameters using data, so its predictions get closer to correct over repeated passes.
The neural network architecture behind most modern language models, introduced in 2017. It's especially good at handling long sequences of text and understanding context.
Training a model on data with no labels, letting it find structure or groupings on its own — such as clustering similar customers together.
Missing a term you'd like explained? Let us know and we'll add it.