AI glossary

AI terms explained in plain English, from chatbots and prompts to neural networks. Each entry has a short definition, an everyday example and links to related terms.

AI basics

Start here: what artificial intelligence is, the main kinds of AI, and the everyday technologies built on it, explained without jargon.

AI literacyAI literacy is knowing enough about how AI works to use it well: when to trust it, how to check it, and where its limits and risks are.AI modelAn AI model is the trained program at the heart of an AI tool: it takes an input, such as a question or a photo, and produces an output.AlgorithmAn algorithm is a step-by-step set of instructions a computer follows to solve a problem or complete a task.Artificial general intelligence (AGI)Artificial general intelligence (AGI) is a hypothetical AI that could learn and reason across almost any task as well as a person can.Artificial intelligence (AI)Artificial intelligence (AI) is technology that lets computers do tasks that normally need human thinking, like understanding language or spotting patterns.Artificial superintelligence (ASI)Artificial superintelligence (ASI) is a hypothetical future AI that would be far more capable than humans at almost everything.Computer visionComputer vision is the field of AI that lets computers understand images and video, such as recognising objects, faces or text in a photo.Facial recognitionFacial recognition is AI that identifies or verifies a person by comparing their face with stored images.Narrow AINarrow AI is AI built for a specific job, such as translating text or recognising faces. All AI that exists today is narrow AI.Natural language processing (NLP)Natural language processing (NLP) is the branch of AI that helps computers understand, interpret and produce human language.Optical character recognition (OCR)Optical character recognition (OCR) is technology that reads text in images or scanned documents and turns it into editable, searchable text.Predictive AIPredictive AI uses patterns in past data to forecast what is likely to happen next, such as which customers might cancel or what a house will sell for.Recommendation systemA recommendation system is AI that suggests things you might like, such as videos, songs or products, based on what you and similar people did before.Rule-based AIRule-based AI makes decisions by following “if this, then that” rules written by people, rather than learning patterns from data.Sentiment analysisSentiment analysis is AI that works out the feeling behind a piece of text, such as whether a review is positive, negative or neutral.Speech recognitionSpeech recognition is AI that turns spoken words into text, so you can talk to a device or get a written transcript of audio.Turing testThe Turing test, proposed by Alan Turing in 1950, asks whether a machine can hold a text conversation that people can’t tell apart from a human’s.

Generative AI and LLMs

How tools like ChatGPT write, reason and make things up: large language models, tokens, context windows, hallucinations and more.

AI hallucinationAn AI hallucination is when an AI confidently gives an answer that is false or made up, such as a fake fact, quote or source.CompletionA completion is the text an AI model writes in response to a prompt. The term comes from models “completing” the text they are given.Context windowThe context window is how much text an AI model can consider at once, including your messages, files and its own replies, measured in tokens.DeepSeekDeepSeek is a Chinese AI company known for open-weight language and reasoning models, and for its own chatbot app.DistillationDistillation trains a smaller AI model to copy the behaviour of a larger one, so you get much of the quality at a fraction of the size and cost.Fine-tuningFine-tuning is extra training that adapts an existing AI model to a specific task, style or field, using a smaller, focused set of examples.Foundation modelA foundation model is a large AI model trained on broad data that can be adapted to many different tasks, from chat to coding to image analysis.Frontier modelA frontier model is one of the most advanced AI models available at a given time, the kind that pushes the limits of what AI can do.Generative AI (GenAI)Generative AI is AI that creates new content, such as text, images, music, video or code, from a description you give it.GPTGPT (generative pre-trained transformer) is OpenAI’s family of large language models, and the technology behind ChatGPT.GroundingGrounding means tying an AI’s answer to specific, trusted sources, such as search results or your documents, instead of relying only on what it memorised.Knowledge cutoffA knowledge cutoff is the approximate date after which an AI model has no information, because its training data stops there.Large language model (LLM)A large language model (LLM) is an AI model trained on vast amounts of text to understand and generate language. It powers chatbots like ChatGPT.LlamaLlama is a family of large language models from Meta whose weights can be downloaded, so developers can run and adapt them on their own hardware.Mixture of experts (MoE)Mixture of experts (MoE) is a model design that splits a large model into many specialised parts and only uses a few of them for each piece of text.Multimodal AIMultimodal AI can understand or create more than one type of content, such as text, images, audio and video, in the same conversation.Pre-trainingPre-training is the first, biggest stage of training an AI model, where it learns general patterns from a huge dataset before being adapted for specific uses.Reasoning modelA reasoning model is an AI model that works through a problem step by step before answering, which helps with maths, coding and complex questions.Reinforcement learning from human feedback (RLHF)Reinforcement learning from human feedback (RLHF) trains an AI model to give better answers by learning from people’s ratings of its responses.Small language model (SLM)A small language model (SLM) is a compact language model with far fewer parameters than an LLM, built to be fast, cheap and able to run on a phone or laptop.TokenA token is a small chunk of text, often a word or part of a word, that a language model reads and writes. Usage limits and prices are counted in tokens.TransformerA transformer is the neural network design behind almost every modern language model. It reads a whole passage at once and works out how its words relate.

Chatbots and assistants

The AI assistants people use every day, from ChatGPT, Gemini and Claude to voice assistants and AI search, and the features they share.

AI assistantAn AI assistant is an AI tool that helps you get things done, such as answering questions, writing, planning or managing tasks, usually through chat or voice.AI companionAn AI companion is a chatbot designed for ongoing personal conversation, friendship or role-play rather than for getting tasks done.AI memoryAI memory is a feature that lets an assistant remember details about you across conversations, such as your name, preferences or ongoing projects.AI OverviewsAI Overviews are the AI-written summaries Google shows at the top of some search results, with links to the pages they draw on.AI search engineAn AI search engine answers your question directly in a written summary built from web pages, instead of only showing a list of links.ChatbotA chatbot is a program you talk to in text or voice that replies like a person would. Modern AI chatbots like ChatGPT are powered by large language models.ChatGPTChatGPT is OpenAI’s AI chatbot. You type or speak a request and it answers, writes, explains, analyses files or creates images.ClaudeClaude is a family of AI models and an AI assistant made by Anthropic, used for writing, analysis, coding and working with long documents.Conversational AIConversational AI is technology that lets computers hold natural back-and-forth conversations with people, in text or speech.Custom GPTA custom GPT is a version of ChatGPT set up for one purpose, with its own instructions, files and tools, that you can build without coding and share.GeminiGemini is Google’s family of AI models and the name of its AI assistant app, which replaced Google Bard in 2024.Microsoft CopilotMicrosoft Copilot is Microsoft’s AI assistant, available as a chat app and built into Windows, Edge and Microsoft 365 apps like Word, Excel and Outlook.PerplexityPerplexity is an AI search engine that answers questions in plain language and lists the web sources it used, so you can check them.Voice assistantA voice assistant is an AI assistant you control by speaking, such as Siri, Alexa or Google Assistant, that answers questions and controls devices.

Prompts and settings

How to ask AI for what you want: prompts, system prompts, few-shot and chain-of-thought prompting, plus settings like temperature and top-p.

Chain-of-thought prompting (CoT)Chain-of-thought prompting asks an AI to work through a problem step by step before giving its answer, which often improves accuracy on harder questions.Context engineeringContext engineering is choosing what an AI model sees for each request, such as instructions, documents, memory and tool results, so it can answer well.Custom instructionsCustom instructions are standing preferences you give a chatbot, like your job or preferred style, so it applies them to every conversation.Few-shot promptingFew-shot prompting means including a few worked examples in your prompt so the AI copies the pattern, format or style you want.Frequency penaltyFrequency penalty is a setting that discourages an AI from repeating the same words, with a stronger effect the more often a word has already appeared.Max tokensMax tokens is a setting that caps how long an AI’s reply can be, measured in tokens. When the limit is reached, the answer stops, even mid-sentence.Presence penaltyPresence penalty is a setting that nudges an AI toward new words and topics by penalising any word that has already appeared, however many times.PromptA prompt is the instruction or question you give an AI tool, such as a message to a chatbot or a description for an image generator.Prompt chainingPrompt chaining breaks a big task into a series of smaller prompts, where each step’s answer feeds into the next.Prompt engineeringPrompt engineering is the skill of writing and refining prompts so an AI gives more accurate, useful and consistent results.Prompt templateA prompt template is a reusable prompt with blanks to fill in, so you get consistent results for a task you do often.Role promptingRole prompting tells the AI to answer as a particular person or expert, such as “Act as a careers coach”, to shape its tone, focus and detail.System promptA system prompt is a set of background instructions that shapes how an AI behaves in every reply, such as its role, tone and rules. Users often don’t see it.TemperatureTemperature is a setting that controls how random or creative an AI’s output is. Low values give focused, predictable answers; high values, more varied ones.Top-pTop-p is a setting that limits an AI’s word choices to its most likely options, such as those making up 90% of the probability. Lower values are more focused.Zero-shot promptingZero-shot prompting means asking an AI to do a task without giving it any examples, relying only on your instructions.

Images, video, audio and writing

The words behind creative AI tools: image generators, diffusion models, text to video, voice cloning, AI writing assistants and more.

AI avatarAn AI avatar is a digital character, often a realistic person, that AI animates to speak a script, used for training videos, marketing and social media.AI humanizerAn AI humanizer rewrites AI-generated text so it sounds more natural and less like a chatbot, often to avoid detection by AI content detectors.AI image generatorAn AI image generator creates pictures from a written description. You type what you want to see, and it produces new images in seconds.AI music generatorAn AI music generator creates songs or instrumental tracks from a text description, sometimes with vocals and lyrics.AI upscalingAI upscaling increases the resolution of an image or video and adds believable detail, so small or blurry pictures look sharper when enlarged.AI writing assistantAn AI writing assistant helps you draft, rewrite, shorten, translate or proofread text, from emails and essays to marketing copy.DALL·EDALL·E is the name of OpenAI’s earlier image-generation models, which create images from text descriptions and were built into ChatGPT.Diffusion modelA diffusion model is a type of generative AI that creates images, video or audio by starting from random noise and gradually refining it into a clear result.InpaintingInpainting is AI image editing that fills or replaces a selected area of a picture, such as removing an object or changing a shirt’s colour.LoRALoRA (low-rank adaptation) is a lightweight way to fine-tune an AI model by training a small add-on file instead of changing the whole model.MidjourneyMidjourney is a popular AI image generator known for its artistic, polished style. You describe an image and it creates several versions to choose from.Negative promptA negative prompt tells an AI image generator what you don’t want in the picture, such as “blurry, extra fingers, text”.SoraSora is OpenAI’s AI video generator, which creates realistic video clips, with sound in newer versions, from text descriptions or images.Stable DiffusionStable Diffusion is an image-generation model from Stability AI whose weights are publicly available, so people can run and customise it on their own computers.Text to speech (TTS)Text to speech (TTS) is technology that reads written text aloud in a natural-sounding voice. AI has made these voices far more lifelike.Text to videoText to video is generative AI that creates short video clips from a written description, and often from a starting image too.Voice cloningVoice cloning is AI that learns to imitate a specific person’s voice from a sample recording, then says new things in that voice.

Agents and automation

What AI agents are, how they use tools and connect to apps, and the automation ideas around them, from MCP to vibe coding.

Agentic AIAgentic AI describes AI that can act with some independence, making decisions and taking several steps toward a goal without a prompt for each one.AI agentAn AI agent is an AI system that works toward a goal on its own, planning steps, using tools like search or apps, and checking its progress along the way.AI automationAI automation uses AI to handle repetitive work automatically, such as sorting emails, extracting data from documents or drafting replies.AI coding assistantAn AI coding assistant helps programmers write, explain, fix and test code, either by suggesting lines as they type or by completing whole tasks as an agent.Computer useComputer use is when an AI agent operates a computer or web browser like a person does: looking at the screen, clicking, typing and scrolling.Deep researchDeep research is a feature in AI assistants where an agent spends several minutes searching many sources and writes a detailed, cited report on your question.Function callingFunction calling lets an AI model ask an app to run a specific action, like checking the weather or searching a database, and then use the result in its reply.Human in the loop (HITL)Human in the loop means a person reviews, approves or corrects an AI system’s work at key points, instead of letting it act entirely on its own.Model Context Protocol (MCP)The Model Context Protocol (MCP) is an open standard for connecting AI assistants to other apps, files and data, so one connector works across many AI tools.Multi-agent systemA multi-agent system is a setup where several AI agents, each with its own role, work together on a task, such as one researching and another writing.No-code AINo-code AI tools let you build AI apps, chatbots and automations with visual editors and plain-language instructions instead of programming.Robotic process automation (RPA)Robotic process automation (RPA) is software that copies the clicks and keystrokes a person makes to complete repetitive computer tasks.Vibe codingVibe coding is building software by describing what you want in plain language and letting AI write and fix the code, often without reading it closely.

Machine learning

How machines learn from data: training, neural networks, supervised and reinforcement learning, overfitting and the building blocks of deep learning.

Activation functionAn activation function decides how strongly each unit in a neural network passes its signal on, which lets the network learn complex, non-linear patterns.Attention mechanismAn attention mechanism lets an AI model focus on the most relevant parts of its input, such as which earlier words matter for understanding the current one.BackpropagationBackpropagation is how a neural network learns from mistakes: it works backwards from the error to find how much each connection caused it, then adjusts them.ClassificationClassification is a machine learning task where the model sorts things into categories, such as “spam” or “not spam”, or which animal is in a photo.ClusteringClustering is a machine learning technique that groups similar items together without being told the groups in advance.Convolutional neural network (CNN)A convolutional neural network (CNN) is a type of neural network designed for images. It scans a picture in small patches to detect edges, shapes and objects.Data augmentationData augmentation creates extra training examples by changing existing ones, such as flipping, cropping or recolouring photos, to help a model generalise.DatasetA dataset is an organised collection of data, such as a table of house prices or a folder of labelled photos, used to train or test AI models.Deep learningDeep learning is a type of machine learning that uses neural networks with many layers to learn complex patterns in text, images and sound.Dimensionality reductionDimensionality reduction simplifies data with many features into fewer, while keeping the most important information, so it is easier to analyse or visualise.Federated learningFederated learning trains an AI model across many devices, such as phones, without collecting their data in one place. Only model updates are shared.Generative adversarial network (GAN)A generative adversarial network (GAN) is two neural networks that compete: one creates fake images and the other tries to spot them, so both keep improving.Gradient descentGradient descent is the method most AI models use to learn: it repeatedly nudges the model’s settings in the direction that reduces its errors.HyperparameterA hyperparameter is a setting chosen before training an AI model, such as how fast it learns or how many layers it has, that shapes how training goes.Loss functionA loss function measures how wrong an AI model’s predictions are during training. Training tries to make this number as small as possible.Machine learning (ML)Machine learning (ML) is a way of building AI where computers learn patterns from examples, instead of being given step-by-step rules.Neural networkA neural network is an AI model made of layers of connected units, loosely inspired by the brain, that learns by adjusting the strength of its connections.OverfittingOverfitting is when an AI model learns its training examples too closely, including their noise, so it does well on them but poorly on new data.ParametersParameters are the internal numbers an AI model learns during training. A model’s size, such as “70 billion parameters”, counts them.Recurrent neural network (RNN)A recurrent neural network (RNN) is a type of neural network that reads data in sequence, one step at a time, keeping a memory of what came before.RegressionRegression is a machine learning task where the model predicts a number, such as a house price, tomorrow’s temperature or next month’s sales.Reinforcement learning (RL)Reinforcement learning (RL) trains an AI by trial and error: it takes actions, gets rewards or penalties, and learns which choices lead to the best results.Self-supervised learningSelf-supervised learning is when an AI model creates its own practice questions from raw data, like hiding a word in a sentence and learning to predict it.Supervised learningSupervised learning trains an AI model on examples that come with the right answer, such as photos labelled “cat” or “dog”, so it can label new examples.Synthetic dataSynthetic data is artificial data generated by computers, often by AI, to train or test models when real data is scarce, expensive or private.Training dataTraining data is the set of examples, such as text, images or past records, that an AI model learns from. Its quality and range shape what the model can do.Transfer learningTransfer learning reuses what an AI model learned on one task as the starting point for a related task, so the new task needs much less data and training.Unsupervised learningUnsupervised learning finds patterns and groups in data that has no labels, such as discovering types of customers from their shopping habits.

Building and running AI

What happens behind an AI app: APIs, inference, GPUs, embeddings, vector databases, RAG, open-source models and running AI on your own device.

AI benchmarkAn AI benchmark is a standard test used to measure and compare how well AI models perform on a task, such as maths problems, coding or general knowledge.AI evalsAI evals are repeatable tests that check whether an AI feature gives good results, using set inputs and clear scoring rules.AI wrapperAn AI wrapper is an app built on top of someone else’s AI model, adding its own interface, prompts or workflow rather than its own model.APIAn API (application programming interface) lets one program use another’s features. AI companies offer APIs so developers can build their models into apps.EmbeddingAn embedding is a list of numbers that represents the meaning of a piece of text, an image or another item, so a computer can measure how similar things are.GPUA GPU (graphics processing unit) is a chip that runs many calculations at once. Built for graphics, GPUs are now the main hardware for training and running AI.InferenceInference is when a trained AI model is used to produce an answer, such as a chatbot writing a reply, as opposed to the training stage when it learns.Local AILocal AI means running an AI model on your own phone or computer instead of on a company’s servers, so it can work offline and keep your data on your device.NPUAn NPU (neural processing unit) is a chip built to run AI tasks efficiently on phones and laptops, without draining the battery or sending data to the cloud.Open-source AIOpen-source AI means models that are free to download, use and modify. Many are “open-weight”: the trained model is shared, but not the data it was trained on.QuantizationQuantization shrinks an AI model by storing its numbers with less precision, so it uses less memory and runs faster, with a small loss in quality.Retrieval-augmented generation (RAG)Retrieval-augmented generation (RAG) makes AI answers more accurate by first looking up relevant information, then having the model answer using it.Semantic searchSemantic search finds results based on what you mean rather than the exact words you type, using AI to understand the intent behind a query.Vector databaseA vector database stores embeddings and quickly finds the ones most similar to a query, which lets AI apps search by meaning across large collections.

Ethics, safety and privacy

The risks and responsibilities that come with AI: bias, deepfakes, privacy, prompt injection, alignment, regulation and how to use AI with care.

AI alignmentAI alignment is the challenge of making an AI system’s goals and behaviour match what people actually intend and value.AI and copyrightAI and copyright covers two open questions: whether training AI on copyrighted work is allowed, and who, if anyone, owns what AI creates.AI biasAI bias is when an AI system produces results that are unfair to certain groups of people, usually because of skewed training data or design choices.AI content detectorAn AI content detector is a tool that estimates whether text or images were made by AI. Its results are guesses, not proof.AI ethicsAI ethics is the study of how AI should be designed and used so it is fair, safe, transparent and respects people’s rights.AI jailbreakAn AI jailbreak is a prompt designed to trick a chatbot into ignoring its safety rules and producing content it is meant to refuse.AI privacyAI privacy is about how AI tools collect, store and use your personal information, including what you type into chatbots and whether it is used for training.AI regulationAI regulation is the set of laws and rules governments create to manage the risks of AI, such as the EU AI Act.AI safetyAI safety is the work of making sure AI systems behave reliably and don’t cause harm, from everyday mistakes to serious misuse and long-term risks.AI slopAI slop is low-quality AI-generated content, such as generic articles, fake images or spammy videos, mass-produced to fill feeds and search results.AI sycophancyAI sycophancy is when a chatbot tells you what you want to hear, agreeing with you or praising your ideas, even when that makes its answer less accurate.AI washingAI washing is exaggerating or misrepresenting how much a product uses AI, or what its AI can do, to make it sound more advanced.AI watermarkingAI watermarking marks content as AI-generated, using visible labels or hidden signals, so people and platforms can tell where it came from.AnthropomorphismAnthropomorphism is treating an AI as if it had human feelings, thoughts or intentions, because it talks like a person.Automation biasAutomation bias is the tendency to trust a computer’s answer more than we should, and to check it less carefully than we would a person’s.DeepfakeA deepfake is a fake video, image or audio clip made with AI to show a real person saying or doing something they never did.Explainable AI (XAI)Explainable AI (XAI) is the effort to make AI decisions understandable to people, so you can see why a system gave a particular answer.GuardrailsGuardrails are checks and limits built into an AI product to stop harmful, off-topic or unauthorised outputs and actions.Prompt injectionPrompt injection is an attack where hidden instructions in content an AI reads, such as a web page or email, trick it into doing something it shouldn’t.Red teamingRed teaming is deliberately attacking an AI system, the way an adversary would, to find weaknesses before it is released.Responsible AIResponsible AI is the practice of building and using AI in ways that are safe, fair, transparent and accountable, with people kept in control.

New to AI? Start here

The terms you’ll meet first when using AI tools.

Featured

See all

Token360One API for 80+ frontier AI models across text, image, audio and video.

AI Overviews TrackerTrack website mentions and citations across six AI engines

AI Image HumanizerHumanize AI-generated images while preserving their visible appearance

WasItAIDetect whether an image was AI-generated or camera-captured.

SeedRouterAccess image, video, and language models via one API without monthly subscriptions

CommentWatchDogHide and delete spam comments on Facebook and Instagram ads

Adobe Firefly Background RemoverAI tool that removes photo backgrounds and outputs a transparent PNG — free to use.

Claude Mark RemoverRemove hidden Claude marks and humanize AI-generated text.

Answer Engine OptimizationSee how AI answer engines mention and cite your website.

C2PA RemoverCheck and remove C2PA content credentials

GenjutsuTransform one video into endless variations with AI.