AI Glossary
55 AI terms explained in plain English, with examples. From agents and tokens to RAG and fine-tuning.
A
AI agent
An AI that carries out a multi-step task for you, not just a single answer. It plans the steps, uses tools such as a browser, apps or files, checks the result and keeps going until the job is done or it needs your approval.
Example Ask an agent to find three flights under a budget and add the best one to your calendar.
AI credits
Units many AI tools use to measure how much you use them. Each action (an image, a minute of voice, a long answer) uses some credits; plans include a monthly amount, and some tools let unused credits roll over.
Example A video tool may charge more credits for a 4K clip than for a 720p one.
Algorithm
A set of step-by-step rules a computer follows to solve a problem. In AI, algorithms decide how a model learns from data and how it turns your input into an output.
Example A recommendation algorithm picks the next video you see.
API
A way for one piece of software to talk to another. AI companies offer APIs so developers can use their models inside their own apps; API use is usually billed separately from chat subscriptions.
Example An app sends text to a speech API and gets an audio file back.
Artificial general intelligence (AGI)
A hypothetical AI that could learn and do almost any intellectual task a person can. Today's AI tools are strong at specific tasks; whether and when AGI arrives is debated, and there is no agreed test for it.
Artificial intelligence (AI)
Computer systems that do tasks that normally need human intelligence. That includes understanding language, recognising images, making predictions and creating text, images, audio or video.
Example A photo app that finds every picture of your dog uses AI.
B
Benchmark
A standard test used to compare AI models. Companies publish benchmark scores for maths, coding or reasoning; they are useful, but a high score does not guarantee a model suits your task.
Example A coding benchmark checks how many programming problems a model solves.
Bias (in AI)
When an AI system produces unfair or skewed results. It usually comes from unbalanced training data or design choices, and can affect things like hiring tools or image generators.
Example An image model that mostly shows men when asked for 'a doctor'.
C
Chatbot
A program you talk to in normal language, by text or voice. Modern chatbots run on large language models and can answer questions, write and summarise.
Example ChatGPT, Claude and Gemini are chatbots.
Computer vision
AI that understands images and video. It can recognise objects, read text in photos, detect faces or describe what is in a picture.
Example Your phone reading a QR code or translating a sign through the camera.
Context window
How much text an AI model can consider at once. It is measured in tokens and covers your messages, uploaded files and the model's replies; anything beyond it is forgotten or cut.
Example A bigger context window lets you ask questions about a whole book at once.
Copilot
An AI assistant built into another app to help you while you work. The word is used generally, and also as a product name by Microsoft and GitHub.
Example An AI that suggests the next line of code as you type.
D
Deep learning
Machine learning that uses neural networks with many layers. It powers most modern AI, from speech recognition to image generators and chatbots.
Deepfake
AI-made video, image or audio that makes a real person appear to say or do something they did not. Deepfakes can be used for fraud and misinformation; many platforms require AI-generated realistic content to be labelled.
Example A fake video of a celebrity endorsing a product they never used.
Diffusion model
A type of AI model that creates images or video by gradually turning random noise into a picture. Most popular image and video generators use diffusion.
Dubbing (AI)
Translating the speech in a video into another language with AI, often keeping the speaker's voice. Some tools also adjust lip movements to match the new language (lip-sync).
Example A YouTube video in English re-voiced in Hindi and Spanish.
E
Embedding
A list of numbers that represents the meaning of a piece of text, an image or audio. Items with similar meaning get similar numbers, which lets software search by meaning instead of exact words.
Example Searching 'cheap laptop' and finding 'budget notebook computer'.
F
Fine-tuning
Training an existing AI model further on your own examples so it behaves the way you want. It is used to teach a model a style, format or specialist knowledge.
Example A company fine-tunes a model on its support tickets to answer customers in its tone.
Foundation model
A large AI model trained on huge amounts of data that many products are built on. Chatbots, writing tools and coding assistants often run on the same few foundation models.
G
Generative AI
AI that creates new content: text, images, audio, video or code. It learns patterns from existing data and produces new material from a prompt.
Example Writing a poem, designing a logo or composing music from a description.
GPU
A graphics processing unit: the computer chip that trains and runs most AI models. GPUs do many calculations at once, which suits AI; shortages and prices of AI chips are a big part of AI industry news.
Grounding
Making an AI answer from specific, trusted sources instead of only its memory. Grounded answers usually show citations, which makes them easier to check.
Example An assistant that answers only from the PDFs you uploaded.
Guardrails
Rules and filters that stop an AI from producing harmful or unwanted output. They can block dangerous instructions, private data or off-topic answers.
H
Hallucination
When an AI states something false as if it were true. Language models predict likely words, so they can invent facts, quotes or sources; always check important answers.
Example A chatbot citing a research paper that does not exist.
I
Inference
Running a trained AI model to get an answer. Training builds the model once; inference happens every time you use it, and it is what most AI usage costs are based on.
L
Large language model (LLM)
An AI model trained on vast amounts of text to understand and generate language. LLMs power chatbots, writing assistants and coding tools.
Example GPT, Claude and Gemini are families of LLMs.
Lip-sync
Matching mouth movements in a video to a voice track. AI video tools use it for avatars and for dubbing into other languages.
M
Machine learning
A way of building AI where the computer learns patterns from examples instead of following hand-written rules. Most AI today is built with machine learning.
Example A spam filter that learns from emails you mark as spam.
MCP (Model Context Protocol)
An open standard that lets AI assistants connect to apps and data in a consistent way. With MCP, an assistant can use tools such as calendars, files or business apps without a custom integration for each.
Model
The trained AI system that turns an input into an output. Products often let you choose between models: faster and cheaper ones, or slower ones that reason more carefully.
Multimodal AI
AI that works with more than one kind of input or output, such as text, images, audio and video. A multimodal assistant can read a photo of a receipt and answer questions about it.
N
Natural language processing (NLP)
The field of AI that deals with understanding and producing human language. It covers translation, summarising, sentiment analysis and chatbots.
Neural network
A computer system loosely inspired by the brain, made of layers of connected units that learn from data. Neural networks are the building blocks of deep learning.
No-code
Building software or automations without writing code, using visual editors. No-code AI tools let anyone connect apps or build simple AI assistants.
Example Connecting a form to a spreadsheet and an email in a drag-and-drop editor.
O
Open-weight model
An AI model whose trained parameters (weights) are published so anyone can download and run it. Some are fully open source; others have licences with limits. They can run on your own computer or server.
P
Parameters
The internal values an AI model learns during training. Models are often described by their number of parameters, in billions; more is not always better for a given task.
Prompt
The instruction or question you give an AI. Clear prompts that state the goal, context, format and audience usually get better results.
Example 'Summarise this article in 5 bullet points for a busy manager.'
Prompt engineering
Writing and refining prompts to get better, more reliable AI output. It includes giving examples, setting a role, asking for a format and breaking big tasks into steps.
R
RAG (retrieval-augmented generation)
A technique where the AI first looks up relevant documents, then writes its answer using them. RAG keeps answers up to date and lets a model use private company knowledge without retraining.
Example A help-desk bot that searches the product manual before answering.
Reasoning model
An AI model that works through a problem step by step before answering. Reasoning models are slower and use more compute, but tend to do better at maths, coding and planning.
RLHF
Reinforcement learning from human feedback: training an AI using people's ratings of its answers. It helps make chatbots more helpful and safer.
S
Speech to text
AI that turns spoken audio into written text (transcription). It powers meeting notes, captions and voice typing.
Example Automatic captions on a video.
Synthetic data
Artificially generated data used to train or test AI. It can fill gaps where real data is scarce or private.
T
Temperature
A setting that controls how predictable or creative an AI's output is. Low temperature gives focused, consistent answers; high temperature gives more varied ones.
Text to image
Creating a picture from a written description. You describe the subject, style and mood, and the model generates images.
Example 'A watercolour of a lighthouse at sunrise.'
Text to speech (TTS)
AI that reads written text aloud in a natural-sounding voice. Used for voiceovers, audiobooks, accessibility and voice assistants.
Text to video
Creating a video clip from a written description or an image. Clips are usually short; longer videos are built by combining clips.
Token
A small chunk of text that AI models read and write: often a word or part of a word. Usage limits, context windows and API prices are measured in tokens.
Example In English, 1,000 tokens is roughly 750 words.
Training data
The examples an AI model learns from. Its quality, size and licensing shape what the model knows and how it behaves.
Transformer
The neural network design behind most modern language models. It pays 'attention' to how words relate across a whole passage, which made today's chatbots possible.
U
V
Vector database
A database that stores embeddings so it can find items by meaning. It is a common part of RAG systems.
Voice cloning
Creating an AI voice that sounds like a specific person from samples of their speech. Reputable tools require consent from the voice owner.
Example A creator cloning their own voice to narrate videos faster.
W
Workflow automation
Setting up apps so that one event automatically triggers actions in others. AI steps can now sit inside workflows, for example summarising an email before saving it.
Example A new form entry adds a row to a sheet and sends a thank-you email.
Z
Zero-shot
When an AI does a task without being shown any examples first. Few-shot means giving a few examples in the prompt, which often improves results.