Last updated: 2026-07-05

AI Glossary

Artificial intelligence has its own vocabulary, and most of it sounds more complicated than it is. Here are 102 of the terms and tools you will run into most often, explained in plain English. Type in the box to filter, or jump by letter.

Agentic AI: A style of AI that acts with a degree of autonomy — setting sub-goals, using tools and adapting — rather than responding once and stopping. AI agents directory ›
AGI (Artificial General Intelligence): A hypothetical AI that can match or exceed human ability across virtually any intellectual task, rather than being good at just one. Today's systems are narrow by comparison.
AI Agent: Software that can plan and carry out multi-step tasks on its own — calling tools, browsing the web, writing code — instead of only answering a single prompt. Browse AI agents ›
AI Assistant: A general-purpose AI you chat with to get help writing, coding, researching or planning. ChatGPT, Claude and Gemini are examples. Compare chatbots ›
AI Safety: The field concerned with making AI systems reliable, controllable and beneficial — spanning technical alignment, testing, red teaming and policy.
Alignment: The effort to make an AI system's behaviour match human intentions and values, so it does what people actually want and avoids harmful actions.
API (Application Programming Interface): A way for one piece of software to talk to another. AI providers offer APIs so developers can build models into their own apps. AI APIs ›
Artificial Intelligence (AI): Computer systems that perform tasks normally needing human intelligence, such as understanding language, recognising images or making decisions. Free AI courses ›
ASI (Artificial Superintelligence): A hypothetical AI that vastly surpasses the best human minds across essentially every domain — a step beyond AGI.
Attention: The transformer mechanism that lets a model weigh which parts of the input matter most when producing each piece of output.
Backpropagation: The core training algorithm that adjusts a neural network's weights by working the error backwards from output to input.
Benchmark: A standard test used to measure and compare how well AI models perform on tasks like reasoning, coding or maths.
Bias: Systematic skew in an AI's outputs, usually reflecting imbalances in its training data. It can lead to unfair or inaccurate results for some groups.
Chain-of-Thought: A prompting approach where the model reasons step by step before answering, which tends to improve accuracy on harder problems. Prompt library ›
Chatbot: A program you converse with in natural language. Modern AI chatbots are powered by large language models. AI chatbots directory ›
ChatGPT tool: OpenAI's flagship AI assistant, built on its GPT models. It handles writing, coding, analysis, image generation and web browsing. Compare chatbots ›
Chunking: Splitting long documents into smaller passages so they can be embedded and retrieved, a key step in retrieval-augmented generation.
Claude tool: Anthropic's AI assistant family, known for writing quality, long-context handling and coding. Compare chatbots ›
Computer Vision: The field of AI concerned with getting computers to interpret images and video — detecting objects, reading text, recognising faces and more.
Context Window: The amount of text a model can consider at once, measured in tokens. A larger context window lets it handle longer documents and conversations.
Copilot: An AI assistant embedded in a tool (an editor, browser or office app) that suggests and helps as you work, rather than a standalone chatbot.
Cursor tool: An AI-first code editor with an agent mode that can edit across a whole codebase. Code generation tools ›
DALL-E tool: OpenAI's text-to-image model, available inside ChatGPT. AI art generators ›
Deep Learning: A branch of machine learning that uses many-layered neural networks to learn complex patterns from large amounts of data.
Deep Research: An AI workflow that reads many sources and synthesises a cited, structured report, rather than giving a quick one-line answer. Research prompts ›
Deepfake: Synthetic image, audio or video generated by AI to convincingly depict someone saying or doing something they did not.
Diffusion Model: A type of generative model that creates images (and increasingly video) by starting from noise and gradually refining it into a picture. AI art generators ›
Distillation: Training a smaller 'student' model to imitate a larger 'teacher', producing a cheaper model that keeps much of the quality.
ElevenLabs tool: A leading AI text-to-speech and voice-cloning platform. Text-to-speech tools ›
Embedding: A numerical representation of text, images or other data that captures meaning, so a computer can measure how similar two things are.
Few-Shot: Giving a model a handful of examples inside the prompt to show it the pattern you want before asking it to do the task.
Fine-Tuning: Further training a pre-trained model on a specific dataset so it performs better on a particular task, domain or style.
Foundation Model: A large model trained on broad data that can be adapted to many downstream tasks. LLMs and image models are examples. Foundation models ›
Function Calling: A capability that lets a model trigger external tools or code in a structured way, so an agent can take real actions, not just produce text. AI agents ›
Generative AI: AI that creates new content — text, images, audio, video or code — rather than only classifying or predicting from existing data.
GitHub Copilot tool: An AI pair-programmer from GitHub that suggests and writes code inside your editor. Code generation tools ›
Google Gemini tool: Google's multimodal AI assistant and model family, integrated across its products. Compare chatbots ›
Google NotebookLM tool: Google's research assistant that reasons over documents you upload and can generate audio overviews. Research tools ›
GPT (Generative Pre-trained Transformer): A family of large language models built on the transformer architecture. The 'pre-trained' part means it learned from vast text before any task-specific tuning.
GPU (Graphics Processing Unit): A chip good at doing many calculations in parallel. GPUs power most AI training and inference.
Grok tool: xAI's AI assistant, integrated with X and equipped with real-time search. Compare chatbots ›
Guardrails: Rules and filters that constrain what an AI system will do or say, reducing harmful, unsafe or off-topic outputs.
Hallucination: When an AI states something false or made-up as if it were fact. Always verify important claims from a model.
Hermes: An open-source local AI agent and open-weight model family from Nous Research that runs on your own machine and can connect to frontier models, known for strong function-calling. Hermes guide ›
Hugging Face tool: A hub for open-source AI — models, datasets, demos and libraries used across the field. Foundation models ›
Inference: Running a trained model to get an output — for example, generating a reply to your prompt. Distinct from training the model in the first place. Model deployment ›
Jailbreak: A prompt or trick designed to bypass an AI's safety guardrails and get it to produce restricted content.
Knowledge Cutoff: The date after which a model has no built-in knowledge, because its training data stops there. Newer facts need web search or provided context.
KV Cache: A store of a model's intermediate values during generation that avoids recomputation and speeds up producing each next token.
LangChain tool: A widely used framework for building applications and agents on top of language models. Agent frameworks ›
Large Language Model (LLM): An AI model trained on huge amounts of text to understand and generate language. It powers chatbots, coding assistants and more. Foundation models ›
Latency: The delay between sending a request to a model and getting a response. Lower latency means a snappier experience.
Latent Space: The internal, compressed representation where a model organises concepts, with similar ideas positioned close together.
Local Model: An AI model that runs on your own computer or server rather than a provider's cloud, keeping data on your machine.
LoRA (Low-Rank Adaptation): An efficient fine-tuning method that trains a small set of extra weights instead of updating the entire model.
Machine Learning (ML): A field of AI where systems learn patterns from data and improve with experience, instead of being explicitly programmed with rules.
MCP Server: A program that implements the Model Context Protocol to expose one app, dataset or capability — files, GitHub, a database — to an AI agent. MCP servers directory ›
Microsoft Copilot tool: Microsoft's AI assistant across Windows, Microsoft 365 and the web. Compare chatbots ›
Midjourney tool: A popular text-to-image generator known for its distinctive, high-quality artistic style. AI art generators ›
Mixture of Experts (MoE): A model design that routes each input to a few specialised sub-networks instead of the whole network, cutting the compute needed per answer.
Model Context Protocol (MCP): An open standard that connects AI agents to external apps, data and services through a common interface, so an agent can read and act on them safely. MCP directory ›
Multi-Agent System: A setup where several AI agents each handle part of a job and coordinate to complete it, often with a controller directing the work. Agent frameworks ›
Multimodal: A model that works across more than one type of input or output — for example text, images and audio together.
n8n tool: A source-available workflow-automation platform with native AI-agent and LLM nodes. No-code agent builders ›
Natural Language Processing (NLP): The area of AI focused on understanding and generating human language, from translation to summarisation to chat.
Neural Network: A model loosely inspired by the brain, made of layers of connected units that adjust as they learn from data.
Ollama tool: A tool for downloading and running open-source LLMs locally on your own machine. Model deployment ›
Open-Source Model: A model whose code — and often weights — are publicly released, so anyone can inspect, run or adapt it.
Open-Weight Model: A model whose trained weights are downloadable and runnable by anyone, even if the training code or data are not fully published.
Orchestration: Coordinating multiple models, tools or agents so a task flows from step to step, with decisions about order, retries and hand-offs. Agent frameworks ›
Overfitting: When a model learns its training data too closely, including noise, and then performs poorly on new, unseen data.
Parameters: The internal values a model learns during training. More parameters can mean more capacity, though quality depends on much more than size.
Perplexity tool: An AI answer engine that searches the web and returns cited answers. Compare chatbots ›
Pre-training: The first, broad phase of training a foundation model on large general data, before any task-specific fine-tuning.
Prompt: The instruction or question you give an AI. The wording strongly affects the quality of the answer. Prompt library ›
Prompt Engineering: The craft of writing prompts that reliably get useful results, using structure, examples and clear instructions. 175+ prompts ›
Prompt Injection: An attack where hidden instructions in a web page or document trick an AI agent into doing something it should not.
Quantisation: Shrinking a model by storing its numbers at lower precision, so it uses less memory and runs faster, usually with a small quality trade-off.
Reasoning Model: A model designed to 'think' through a problem across multiple steps before answering, trading speed for accuracy on hard tasks.
Red Teaming: Deliberately probing an AI system for weaknesses and harmful behaviours before release, so they can be fixed.
Reinforcement Learning from Human Feedback (RLHF): A training stage where human preferences are used to steer a model's responses toward what people find helpful and safe.
Reranking: A second pass that reorders retrieved results by relevance before they are handed to the model, improving RAG quality.
Retrieval-Augmented Generation (RAG): A technique that fetches relevant documents and feeds them to a model at answer time, so responses are grounded in specific sources. Vector databases ›
Sora tool: OpenAI's text-to-video model, which generates short clips from written descriptions. Text-to-video tools ›
Speech-to-Text (STT): Technology that transcribes spoken audio into written text. Also called speech recognition. Speech tools ›
Stable Diffusion tool: An open-source text-to-image diffusion model that can run on your own hardware. AI art generators ›
System Prompt: Hidden instructions that set an AI assistant's role, tone and rules for a whole conversation, separate from the user's messages.
Temperature: A setting that controls randomness in a model's output. Lower values give focused, predictable answers; higher values give more varied, creative ones.
Test-Time Compute: Extra computation a model spends reasoning at answer time — rather than during training — to improve results on hard problems.
Text-to-Image: Generating a picture from a written description. Midjourney and similar tools work this way. AI art generators ›
Text-to-Speech (TTS): Turning written text into natural-sounding spoken audio. Text-to-speech tools ›
Text-to-Video: Generating video clips from a written description, a fast-moving area of generative AI. Text-to-video tools ›
Token: A chunk of text — a word or part of a word — that models read and generate. Usage and pricing are often measured in tokens.
Tokenisation: Breaking text into tokens so a model can process it. Different models split text in slightly different ways.
Tool Use: A model's ability to call external tools — search, code execution, apps — to do things it cannot do from text alone. AI agents ›
Training: The process of teaching a model by adjusting its parameters on large datasets so it learns useful patterns.
Transformer: The neural-network architecture behind most modern language and multimodal models, notable for its 'attention' mechanism.
Vector Database: A database that stores embeddings and finds items by similarity, a key piece of retrieval-augmented generation. Vector databases ›
Vibe Coding: Building software by describing what you want in plain language and letting an AI generate and iterate on the code. Code generation tools ›
Watermarking: Embedding an invisible signal in AI-generated content so it can later be identified as machine-made.
Weights: The learned parameter values inside a trained model. Releasing weights lets others run the model themselves.
Zero-Shot: Asking a model to do a task with no examples in the prompt, relying purely on what it learned during training.

No terms match your search.

New to AI? Start with our free beginner courses, browse the prompt library, or explore AI agents & MCP.

AI Glossary

A

B

C

D

E

F

G

H

I

J

K

L

M

N

O

P

Q

R

S

T

V

W

Z