Every AI term, translated for humans.

Agentic AI describes AI systems designed to act with a degree of autonomy — setting sub-goals, using tools, making decisions and completing multi-step tasks — rather than only responding to one prompt at a time.

An AI agent is software that uses an AI model to pursue a goal on its own: it plans steps, takes actions with tools (search, code, apps, APIs), observes the results and keeps going until the task is done or it needs a human.

AI alignment is the work of making AI systems pursue the goals, values and intentions their designers and users actually want, rather than unintended or harmful behaviors.

An AI benchmark is a standardised test — a fixed set of tasks or questions with a scoring method — used to measure and compare how well AI models perform at skills like reasoning, coding, math, knowledge or following instructions.

An AI copilot is an AI assistant built into the software you already use — a code editor, word processor, spreadsheet or CRM — that suggests, drafts and completes work alongside you while you stay in control of decisions.

AI guardrails are rules, filters and checks placed around an AI model or application to keep its inputs and outputs safe, accurate, on-topic and compliant — for example blocking harmful content, protecting personal data or requiring human approval for risky actions.

An AI hallucination is when a model generates information that sounds plausible and confident but is false or made up — such as fake facts, quotes, citations, statistics or URLs.

An AI jailbreak is a prompt or technique that tricks a model into ignoring its safety rules and producing content it was designed to refuse.

AI Overviews are AI-generated summaries that Google shows at the top of some search results, answering the query directly and linking to sources. The feature evolved from Google's Search Generative Experience (SGE).

AI safety is the field focused on preventing AI systems from causing harm, whether through accidents, misuse, security failures or misaligned goals, from today's chatbots to hypothetical future superintelligence.

AI slop is low-quality digital content mass-produced with AI — generic articles, uncanny images, spammy videos, fake books and filler posts made for clicks or volume rather than value. Merriam-Webster named 'slop' its 2025 Word of the Year.

An AI wrapper is a product built on top of someone else's AI model — usually accessed through an API — that adds its own interface, prompts, workflow or data for a specific use case, rather than developing its own model.

Answer engine optimization (AEO) is the practice of formatting content so that search features, voice assistants and AI chatbots can pull it out as a direct answer to a user's question.

An API (application programming interface) is a defined way for one piece of software to request data or actions from another. AI APIs let developers send prompts to models like GPT, Claude or Gemini from their own apps and pay per use.

Artificial general intelligence (AGI) is a hypothetical AI system that can learn and perform most intellectual tasks a human can, across domains, rather than excelling only at narrow, specific jobs.

Artificial superintelligence (ASI) is a hypothetical AI that greatly surpasses the best human minds in virtually every domain, including science, strategy, creativity and social skill.

Chain of thought (CoT) is when an AI model writes out intermediate reasoning steps before its final answer, either because it was prompted to or trained to, which often improves accuracy on multi-step problems.

A chatbot is software that holds a conversation with people through text or voice. Modern AI chatbots like ChatGPT, Claude and Gemini use large language models to understand open-ended questions and generate natural replies.

Computer vision is the field of AI that enables computers to interpret and understand images and video — recognising objects, faces, text, scenes and movement, and making decisions based on what they 'see'.

A context window is the maximum amount of text, measured in tokens, that an AI model can consider at once — including your prompt, any documents or conversation history, and the model's own reply.

Deep learning is a type of machine learning that uses neural networks with many layers to learn complex patterns directly from large amounts of raw data such as text, images and audio. It powers modern AI like chatbots, image generators and speech recognition.

A deepfake is AI-generated or AI-altered video, audio or imagery that realistically depicts a real person saying or doing something they never actually said or did.

A diffusion model is a type of generative AI that creates images, video or audio by starting with random noise and gradually removing it, step by step, until it forms an output that matches a prompt.

Embeddings are lists of numbers (vectors) that represent the meaning of text, images or other data, so that items with similar meanings end up close together and a computer can measure how related they are.

Fine-tuning is the process of further training an existing AI model on a smaller, specialised dataset so it performs better at a particular task, style, format or domain.

Generative AI is artificial intelligence that creates new content — text, images, audio, video or code — by learning patterns from large amounts of existing data and producing original output in response to a prompt.

Generative engine optimization (GEO) is the practice of structuring and improving content so that AI systems like ChatGPT, Perplexity, Gemini and Google AI Overviews cite, quote or recommend it in their generated answers.

GPT stands for Generative Pre-trained Transformer, a family of large language models developed by OpenAI that power ChatGPT. The term is also used loosely for any similar text-generating AI model.

Inference is the process of running a trained AI model to produce an output, such as generating a chatbot reply or classifying an image, as opposed to training, which is how the model learns.

A large language model (LLM) is an AI system trained on enormous amounts of text to predict the next word (token) in a sequence, which lets it write, summarize, translate, answer questions and generate code.

LLMOps is the set of practices and tools for deploying, monitoring, evaluating and improving applications built on large language models in production — covering prompts, model versions, costs, quality, safety and data.

Machine learning is a branch of artificial intelligence in which computers learn patterns from data to make predictions or decisions, instead of being explicitly programmed with step-by-step rules for every case.

Mixture of experts (MoE) is an AI model design in which the network contains many specialised sub-networks ('experts') and a router that activates only a few of them for each token, letting a model have huge total capacity while using only part of it per step.

The Model Context Protocol (MCP) is an open standard, introduced by Anthropic in November 2024, that gives AI applications a common way to connect to external tools, data sources and services through 'MCP servers'.

Model distillation is a technique for training a smaller, faster 'student' AI model to imitate the outputs of a larger, more capable 'teacher' model, so it keeps much of the teacher's ability at a fraction of the size and cost.

Multimodal AI is artificial intelligence that can understand and/or generate more than one type of data — such as text, images, audio and video — within a single model or system.

Natural language processing (NLP) is the field of AI focused on enabling computers to understand, interpret and generate human language, in text or speech — from spam filters and translation to chatbots and large language models.

A neural network is a computing system made of layers of simple connected units ('neurons') that learn to recognise patterns by adjusting the strength of their connections based on examples, loosely inspired by the brain.

No-code automation is using visual tools to connect apps and automate repetitive tasks, like moving data, sending emails or calling AI models, without writing code.

Open-weight models are AI models whose trained parameters (weights) are publicly released, so anyone can download, run, fine-tune and often commercially use them, subject to the model's license.

Parameters are the internal numbers, mainly weights, that an AI model adjusts during training and uses to make predictions. A model's size is usually described by its parameter count, such as 7 billion or 70 billion.

Prompt engineering is the practice of writing and structuring instructions for an AI model — context, task, examples, constraints and output format — so it reliably produces the result you want.

Prompt injection is an attack where malicious instructions are hidden in content an AI processes, such as a webpage, email or document, causing the AI to follow the attacker's commands instead of its user's or developer's.

Quantization is a technique that shrinks an AI model by storing its numbers (weights) at lower precision — for example 8-bit or 4-bit instead of 16-bit — so it uses less memory and runs faster, usually with only a small loss in quality.

Reasoning models are AI language models trained to work through a problem step by step, generating internal reasoning before giving a final answer, which improves performance on math, coding, logic and multi-step tasks.

Retrieval-augmented generation (RAG) is a technique where an AI system first searches a set of documents or data for information relevant to a question, then gives that information to a language model so its answer is grounded in those sources.

Synthetic data is artificially generated data — created by algorithms or AI models rather than collected from real-world events — that mimics the patterns of real data and is used for training, testing or sharing without exposing real records.

A system prompt is a set of instructions given to an AI model before the conversation starts, defining its role, tone, rules and constraints for every response that follows.

Temperature is a setting that controls how random an AI model's output is: low temperature makes responses more predictable and focused, high temperature makes them more varied and creative.

Text-to-speech (TTS) is technology that converts written text into spoken audio. Modern AI text-to-speech uses deep learning to produce natural-sounding voices with realistic intonation, emotion and pacing.

Text-to-video is generative AI that creates video clips from a written description (a prompt), often also accepting images or reference footage, by generating a sequence of coherent frames with motion.

The singularity (or technological singularity) is a hypothetical future point where AI becomes able to improve itself so rapidly that technological progress becomes uncontrollable and unpredictable to humans.

The Turing test is a test proposed by mathematician Alan Turing in 1950 in which a human judge holds text conversations with a machine and a person; if the judge can't reliably tell which is the machine, the machine is said to have passed.

A token is the basic unit of text an AI language model reads and writes — usually a word, part of a word, or a punctuation mark. Models measure input, output, context limits and API pricing in tokens.

Training data is the collection of examples, such as text, images, audio or code, that an AI model learns from. Its size, quality and content largely determine what the model knows and how it behaves.

A transformer is a neural network architecture, introduced by Google researchers in the 2017 paper 'Attention Is All You Need', that uses a mechanism called attention to weigh how every part of an input relates to every other part. It underpins nearly all modern large language models.

A vector database stores embeddings — numerical representations of meaning — and can quickly find the items most similar to a query, which makes it a common backbone for semantic search and retrieval-augmented generation (RAG).

Vibe coding is building software by describing what you want to an AI in plain language and accepting the code it writes, often without reading or fully understanding it. The term was coined by Andrej Karpathy in February 2025.

Voice cloning is using AI to create a synthetic copy of a specific person's voice from recordings of them speaking, so a computer can generate new speech that sounds like that person saying words they never actually said.

Workflow automation is using software to carry out a sequence of business tasks automatically — moving data between apps, triggering actions and, increasingly, using AI to read, decide and write — with little or no manual effort.