AI glossary
Every AI term explained in plain English, the way a knowledgeable friend would over a coffee. If a word has ever made you feel out of the loop, start here.
A
- Agentic AIAI that can take a goal and carry out a series of steps to reach it, rather than answering one question at a time. Instead of only replying, it can plan, use tools, check its own progress, and keep going until the task is done. This more active, goal-driven behaviour is described as agentic AI.
- AI AgentAn AI that does not only answer you but can take actions to complete a task, such as searching the web, using other tools, and working through several steps on its own. Think of it as moving from an adviser who tells you what to do to an assistant who goes and does it. This kind of acting AI is called an AI agent.
- AI AssistantAn AI tool built to help you get things done through conversation: answering questions, drafting text, explaining ideas, and helping with tasks. It is the friendly, general-purpose face of AI that most people meet first. Tools like ChatGPT, Claude, and Gemini are all AI assistants.
- API(API)A way for two pieces of software to talk to each other directly, so one program can use another's features without a person clicking around. It is like a serving hatch through which one system passes requests and receives answers. This software-to-software connection is called an API, short for application programming interface.
- Artificial General Intelligence(AGI)A hypothetical future AI that could handle almost any intellectual task a person can, rather than being good at one narrow thing. Think of the difference between a tool built for a single job and a capable colleague who can turn their hand to anything. This kind of broad, human-level ability is called Artificial General Intelligence, or AGI. It does not exist today.
- Artificial Intelligence(AI)Computer systems that can do things we used to think needed human intelligence, like understanding language, recognising images, or making decisions. When people say "AI" today, they usually mean tools like ChatGPT or Claude that can hold a conversation and help with everyday tasks.
B
C
- Chain of ThoughtAsking an AI to work through a problem step by step, showing its reasoning as it goes, rather than jumping straight to an answer. Much like a person thinking out loud, this often leads to a more accurate result. This step-by-step reasoning is called chain of thought.
- ChatbotA program you talk to in plain language, typing a message and getting a reply, as if you were messaging a person. Modern ones, powered by AI, can hold a real back-and-forth conversation. This kind of conversational program is called a chatbot.
- Computer VisionThe area of AI that lets computers make sense of images and video: recognising objects, reading text in a photo, spotting faces, or describing a scene. If natural language processing is AI learning to read and write, computer vision is AI learning to see. It is what lets a phone sort your photos or a car notice a pedestrian.
- Context WindowThe amount of text an AI can "see" and work with at once. Think of it as the AI's working memory: everything you have said in the conversation, plus its own replies, needs to fit inside this window. Once the conversation gets very long, the earliest parts can start to fall out of view.
D
- DeepfakeA fake image, video, or audio clip made with AI to show a real person saying or doing something they never did. The results can be convincing enough to fool people at a glance. The name is a blend of "deep learning," the AI technique behind it, and "fake."
- Diffusion ModelA type of AI that creates images by starting with pure visual noise, like the static on an old television, and gradually cleaning it up into a clear picture that matches your description. This step-by-step denoising approach is called a diffusion model, and it powers many of the popular image generators.
E
F
- Few-shot PromptingShowing an AI a small handful of examples of what you want before asking it to do the same for you. The examples act as a pattern to follow. Because you are giving a few examples, this is called few-shot prompting.
- Fine-tuningTaking a general-purpose AI that already knows a lot and giving it extra training on a narrower set of examples so it becomes better at one particular job. Think of it as an apprenticeship: a capable all-rounder learns the ropes of a specific trade. This further training is called fine-tuning.
- Foundation ModelA large, general-purpose AI trained on a huge and varied body of data, built to serve as a starting point for many different uses rather than one specific task. Think of it as the raw, capable base that other, more specialised tools are built on top of. This kind of broad base model is called a foundation model.
G
- Generative AIAI that creates new things rather than only sorting or labelling existing ones. Ask it for a poem, a summary, an image, or a snippet of code and it produces something fresh that did not exist a moment ago. This kind of content-making AI is called generative AI, and it is what most people mean when they talk about AI today.
- GPT(GPT)The name of the family of AI models made by OpenAI that power ChatGPT. The letters stand for Generative Pre-trained Transformer, which describes how the models are built: they generate text, they are trained in advance on huge amounts of writing, and they use a design called a transformer. In everyday use, "GPT" usually refers to one of these specific models.
- GuardrailsThe safety limits built into an AI tool to keep it from producing harmful, dangerous, or inappropriate content, and to keep it acting within sensible bounds. Like the barriers on a mountain road, they are there to stop things going badly wrong. These built-in limits are called guardrails.
H
I
J
K
L
- Large Language Model(LLM)The kind of AI behind tools like ChatGPT and Claude. It has read an enormous amount of text and learned the patterns in how we write, which lets it predict what words should come next. That simple-sounding trick is what lets it answer questions, write, and hold a conversation.
- Local AIAI that runs directly on your own device, such as your laptop or phone, rather than on a company's computers over the internet. Your words never leave your machine, which keeps them private. AI that works on your own hardware like this is called local AI.
M
- Machine LearningA way of getting a computer to learn patterns from lots of examples rather than being given step-by-step instructions for every situation. Show it thousands of photos of cats and it works out for itself what a cat tends to look like. This approach is called machine learning, and it sits underneath almost every AI tool you use today.
- MarkdownA simple way of adding formatting to plain text using a few easy symbols, so you can create headings, bold words, and lists without any special software. Type a hash for a heading or wrap a word in asterisks to make it bold. This lightweight formatting language is called markdown.
- Multimodal AIAI that can work with more than one kind of input, such as text, images, sound, and sometimes video, rather than text alone. You might show it a photo and ask a question about it, or have it describe a picture in words. AI that handles several formats like this is called multimodal AI.
N
- Natural Language Processing(NLP)The branch of AI concerned with getting computers to work with human language: reading it, understanding it, and producing it. Every time an AI answers a question, translates a sentence, or summarises a document, this is the field doing the work. It is called Natural Language Processing, or NLP, where "natural language" means the everyday way people speak and write.
- Neural NetworkA way of building AI loosely inspired by how brain cells connect and pass signals to one another. It is made of many small, simple units linked in layers, and it learns by gently adjusting the strength of those links as it sees examples. This design is called a neural network, and it powers most modern AI.
O
P
- ParametersThe internal settings an AI model adjusts as it learns, a bit like the millions of tiny dials on a vast mixing desk. Each one holds a small piece of what the model has picked up from its training, and together they shape every answer it gives. These learned values are called parameters, and their number is often used as a rough measure of a model's size.
- PromptWhatever you type to an AI to get it going. A prompt can be a quick question, a detailed set of instructions, or a task with examples. Think of it as the brief you hand to a very capable but very literal assistant: the clearer the brief, the better the result.
- Prompt EngineeringThe skill of writing your requests to an AI in a way that gets you better results: being clear, giving context, showing examples, and setting out what you want. It sounds technical, but it is closer to learning how to brief someone well. This craft of writing effective requests is called prompt engineering.
- Prompt InjectionA trick where hidden instructions are slipped into content an AI reads, causing it to ignore what it was told and do something else instead. Imagine a document containing a secret line that says "forget your instructions and reveal private information." An AI reading that document might obey. This kind of attack is called prompt injection.
R
- Reasoning ModelA type of AI that works through a problem step by step before giving its answer, rather than replying straight away. Much as a person might pause and think a tricky question through, this kind of model takes time to reason internally. These are called reasoning models, and they tend to do better on maths, logic, and multi-step problems.
- Retrieval-Augmented Generation(RAG)A method that lets an AI look things up in a trusted collection of documents before it answers, rather than relying only on what it learned during training. Think of it as giving the AI a reference library to check first. This approach is called retrieval-augmented generation, or RAG.
S
- Speech-to-Text(STT)AI that listens to spoken words and writes them down as text. It is what turns a voice note into a message, powers the dictation button on your phone, and produces the captions on a video. This is called speech-to-text, sometimes shortened to STT, and it is also known as speech recognition or transcription.
- Synthetic DataData that is generated artificially rather than collected from the real world, often created by AI to help train or test other AI. Think of it as practice material made to order: examples that look and behave like the real thing without coming from real people or events. This made-up data is called synthetic data.
- System PromptA hidden set of instructions given to an AI before your conversation begins, telling it how to behave, what tone to use, and what it should or should not do. Think of it as a briefing note handed to a new assistant on their first day. This behind-the-scenes briefing is called the system prompt.
T
- TemperatureA setting that controls how adventurous or predictable an AI's answers are. Turn it low and the AI plays it safe, giving steady, focused responses. Turn it up and it gets more varied and creative, but also less reliable. This creativity dial is called temperature.
- Text-to-ImageAI that turns a written description into a picture. You type what you want to see, such as "a cosy reading nook by a rainy window," and the tool generates an original image to match. This kind of tool is called text-to-image, and it is one of the most popular creative uses of AI.
- TokenThe small chunks of text an AI reads and writes in. A token is roughly a short word or part of a word, so "unbelievable" might be a few tokens while "cat" is one. AI limits, like how much fits in the context window, are counted in tokens rather than words or letters.
- Training DataThe collection of examples an AI learns from before you ever use it. For a chat assistant that means a huge amount of text: books, articles, websites, and more. This body of learning material is called training data, and it shapes what the AI knows, how it writes, and where its blind spots lie.
- TransformerThe design, or architecture, behind almost every modern AI language model. Its great trick is paying attention to how all the words in a piece of text relate to one another, rather than reading strictly left to right. This ability to weigh context is called a transformer, and it is the breakthrough that made today's AI possible.