- Home
- Glossary
Reference
AI glossary for working professionals
124 terms, each defined in one clear sentence and then explained with a real workplace example. No math required.
A
- Agentic AIAgentic AI describes AI systems designed to act with autonomy: they take a goal, break it into steps, use tools, and carry the work through with limited human direction. It is the broader approach; an AI agent is a specific system built that way.
- AI agentAn AI agent is a system that uses an AI model to pursue a goal on its own by planning steps, using tools such as web browsers, files, and software APIs, and adjusting based on results, rather than answering a single prompt.
- AI assistantAn AI assistant is a software application, usually built on a large language model, that helps you with tasks through natural conversation: drafting, summarizing, answering questions, analyzing files, and increasingly taking actions in other apps.
- AI avatarAn AI avatar is a synthetic on-screen presenter, either a stock character or a digital likeness of a real person, that speaks any script you type. Tools such as Synthesia and HeyGen use it to turn text into talking-head video without a camera or crew.
- AI biasAI bias is a systematic tendency for an AI system to produce outputs that are unfair or skewed toward certain groups, usually because the data it learned from reflected historical patterns, gaps, or human prejudice.
- AI detectorAn AI detector is a tool that tries to estimate whether a piece of text, an image, or audio was generated by AI. Detectors for text in particular are unreliable, with false positives and easy evasion, and should not be used as sole evidence.
- AI governanceAI governance is the set of policies, roles, controls, and oversight processes an organization uses to decide how AI is adopted, used, monitored, and held accountable across the business.
- AI literacyAI literacy is the practical understanding of what AI systems can and cannot do, how they produce their outputs, where they fail, and how to use them responsibly and effectively in your own work.
- AI memoryAI memory is a feature that lets an AI assistant retain facts across separate conversations, such as your role, preferences, and ongoing projects, so you do not have to repeat them. It differs from the context window, which lasts only for one conversation.
- AI OverviewsAI Overviews are the AI-generated summaries Google shows at the top of many search results, assembled from web sources with citation links. They answer the question directly, so many searchers never click through to the sites the answer came from.
- AI safetyAI safety is the field concerned with preventing AI systems from causing harm, from everyday failures like confident errors and biased outputs to misuse by bad actors and risks from highly capable future systems.
- AI slopAI slop is low-quality, mass-produced content generated by AI and published with little or no human effort, usually to capture clicks, fill a quota, or look busy. The term covers text, images, video, and code that is generic, padded, or subtly wrong.
- AI watermarkingAI watermarking embeds a hidden or attached signal in AI-generated content, such as an invisible pattern in an image or metadata in a file, so that its origin can be identified later. It helps with provenance but is far from foolproof.
- AlignmentAlignment is the work of making an AI system pursue the goals and values its designers and users actually intend, rather than a shortcut, a literal reading of the instructions, or a goal of its own.
- Anomaly detectionAnomaly detection is the automated flagging of data points or events that do not fit the normal pattern: a purchase in an unusual place, an expense claim twice the usual size. It powers fraud alerts, security monitoring, and equipment warnings.
- AnthropomorphismAnthropomorphism is the habit of attributing human qualities such as understanding, intent, or feelings to something that is not human. With AI chatbots, it means reading a fluent, first-person reply as a sign of a mind behind it.
- API (application programming interface)An API (application programming interface) is a way for one piece of software to talk to another. An AI API lets your own tools, spreadsheets, or automations send a prompt to a model and receive the response, without a person typing in a chat window.
- Artificial general intelligence (AGI)Artificial general intelligence (AGI) is a hypothetical AI that could learn and perform any intellectual task a human can, across domains, rather than being built for specific tasks. It does not exist today, and its definition is contested.
- Artificial intelligence (AI)Artificial intelligence (AI) is software that performs tasks that normally need human judgment: understanding language, recognizing images, making predictions, or generating text. Today it usually means systems built with machine learning.
- Attention mechanismThe attention mechanism is the part of a transformer model that lets every word in the input weigh its relationship to every other word, so the model can tell which earlier words matter most when interpreting or generating the next one.
- Automation biasAutomation bias is the human tendency to trust a suggestion from an automated system over one's own judgment or contradicting evidence, even when the system is wrong. With AI, it shows up as accepting fluent, confident output without checking it.
C
- Chain-of-thought promptingChain-of-thought prompting asks an AI model to work through a problem step by step before giving its final answer. Showing its reasoning tends to improve accuracy on math, logic, and multi-step tasks and makes mistakes easier to spot.
- ChatbotA chatbot is software you converse with by typing or speaking. Older chatbots followed scripted decision trees; modern AI chatbots such as ChatGPT, Claude, Gemini, and Copilot use large language models to understand and respond to almost any request.
- Computer useComputer use is an AI capability in which a model operates a computer the way a person does: looking at the screen, moving the cursor, clicking, typing, and switching between apps and web pages to finish a task in software that has no direct integration.
- Computer visionComputer vision is the field of AI that lets software interpret images and video: recognizing objects, reading text, detecting defects, tracking movement, or describing a scene. It powers everything from phone photo search to warehouse inspection cameras.
- Context engineeringContext engineering is the practice of choosing what an AI model sees when it works on a task: instructions, documents, examples, tool results, and history, and how they are arranged. It extends prompt engineering from wording to the whole input.
- Context windowThe context window is the maximum amount of text, measured in tokens, that a language model can consider at one time. It includes your prompt, any pasted documents, the conversation so far, and the model's reply.
- Conversational AIConversational AI is the category of technology that lets software understand and respond to human language in back-and-forth dialogue, typed or spoken. It covers everything from simple website chatbots to voice agents and general-purpose AI assistants.
- CopilotA copilot is an AI assistant built into the software you already use, suggesting, drafting, and completing work alongside you rather than replacing you. Copilot is also Microsoft's brand for its AI assistants across Windows, Microsoft 365, and GitHub.
- Custom GPTA custom GPT is a tailored version of ChatGPT that you configure with your own instructions, uploaded reference files, and optional connections to outside services, then reuse or share without rewriting the setup each time.
D
- Data labelingData labeling is the work of attaching correct answers to raw data, such as tagging emails as spam, drawing boxes around pallets in photos, or rating chatbot replies, so a machine learning model can learn from those examples.
- Data privacyData privacy, in the context of AI, is the set of rules and practices that govern what information you put into an AI system, who can see it, how long the vendor keeps it, and whether it is used to train future models.
- Deep learningDeep learning is machine learning that uses neural networks with many layers, which lets a model learn complex patterns directly from raw data such as text, images, or audio. It is the technique behind today's language models and image generators.
- DeepfakeA deepfake is synthetic audio, video, or imagery, generated or altered by AI, that convincingly depicts a real person saying or doing something they never did.
- Diffusion modelA diffusion model is a type of generative AI that creates images, audio, or video by starting with random noise and removing it step by step until a coherent result matching the prompt emerges. Most image generators of recent years are built this way.
E
- EmbeddingAn embedding is a list of numbers that represents the meaning of a piece of text, an image, or other content, so that items with similar meaning end up with similar numbers. Embeddings are what let AI systems search by meaning instead of exact words.
- ExplainabilityExplainability is the degree to which you can understand why an AI system produced a particular output. It matters most when a decision affects a person and someone has to justify it.
F
- Few-shot promptingFew-shot prompting is including a handful of examples of the input and output you want inside your prompt so the AI model can copy the pattern. It is the fastest way to get consistent formatting, tone, and judgment calls.
- Fine-tuningFine-tuning is extra training that adjusts an existing AI model's weights using a smaller, targeted set of examples so it performs better on a specific task, style, or domain. It changes the model itself, unlike prompting, which only changes the instructions.
- Foundation modelA foundation model is a large AI model trained on broad, general data that can be adapted to many different tasks, rather than built for a single purpose. Most of today's chat assistants, image generators, and coding tools are built on one.
- Frontier modelA frontier model is one of the most capable AI models available at a given time, typically the newest flagship from a major lab. The term is used in policy and industry to describe models whose capabilities may create new risks that need extra scrutiny.
G
- Generative AIGenerative AI is a class of AI models that create new content, including text, images, code, audio, and video, in response to a prompt. Chat assistants like ChatGPT, Claude, and Gemini and image tools like Midjourney are generative AI.
- Generative engine optimization (GEO)Generative engine optimization (GEO) is the practice of shaping your content and online presence so that AI answer engines such as ChatGPT, Perplexity, Google's AI Overviews, and Copilot cite or recommend your business when people ask them questions.
- GPTGPT stands for generative pre-trained transformer, the family of large language models from OpenAI that powers ChatGPT. The name describes how the models are built: they generate text, are pre-trained on huge datasets, and use the transformer architecture.
- GPU (graphics processing unit)A GPU is a processor built to perform enormous numbers of simple calculations in parallel. Originally for graphics, GPUs are now the main hardware used to train and run AI models, which is why they are central to AI cost, speed, and supply.
- Ground truthGround truth is the verified, correct answer that an AI system's output is compared against, such as a human-labeled dataset or a confirmed real-world outcome. It is the reference standard used to train models and measure their accuracy.
- GroundingGrounding means tying an AI model's answer to specific, verifiable sources, such as documents you provide, a database, or live search results, instead of relying on what the model remembers from training.
- GuardrailsGuardrails are the rules, filters, and technical limits placed around an AI system to keep its behavior within acceptable bounds, such as refusing harmful requests, staying on topic, protecting data, or requiring approval before acting.
H
- HallucinationA hallucination is when an AI model produces information that is false or fabricated but presents it confidently, such as an invented statistic, a citation to a paper that does not exist, or a made-up product feature.
- Human-in-the-loopHuman-in-the-loop is a design approach in which a person reviews, approves, or corrects an AI system's output at defined points before it takes effect, combining the speed of automation with human judgment and accountability.
I
- InferenceInference is the act of running a trained AI model to produce an output, such as answering a prompt or classifying an image. Training happens once; inference happens every time anyone uses the model, and it is what you pay for as a user.
- Intelligent document processing (IDP)Intelligent document processing (IDP) uses OCR, machine learning, and language models to pull structured data out of documents such as invoices, contracts, and claims, validate it, and feed it into business systems, with humans reviewing the exceptions.
L
- Large language model (LLM)A large language model (LLM) is a neural network trained on enormous amounts of text to predict the next word, then tuned to follow instructions and hold a conversation. LLMs power ChatGPT, Claude, Gemini, and Microsoft Copilot.
- LatencyLatency is the delay between sending a request to an AI system and receiving its response. For language models it is usually measured as time to the first word and then words per second as the answer streams in.
- Long contextLong context describes an AI model's ability to take in and work with very large inputs, from a whole contract to an entire codebase, in a single request. It is set by the size of the model's context window, measured in tokens.
M
- Machine learning (ML)Machine learning is the branch of AI in which software learns patterns from examples instead of following rules a programmer wrote by hand. Nearly every modern AI tool, including chat assistants, is built on it.
- Mixture of experts (MoE)A mixture of experts (MoE) is a model design that splits the network into many specialized sub-networks called experts and activates only a few of them for each token, giving the capacity of a huge model at the running cost of a much smaller one.
- Model cardA model card is a standardized document published alongside an AI model that describes what it was built to do, how it was trained and tested, where it performs well or poorly, and what its known risks and limitations are.
- Model Context Protocol (MCP)The Model Context Protocol (MCP) is an open standard, introduced by Anthropic in late 2024, that lets AI assistants connect to external tools, data sources, and services through a common interface instead of custom one-off integrations.
- Model distillationModel distillation is a training technique in which a smaller 'student' model learns to imitate the outputs of a larger 'teacher' model, producing a model that is cheaper and faster to run while keeping much of the teacher's capability.
- Model evaluationModel evaluation is the process of measuring how well an AI model performs on a defined set of tasks, using test data, benchmarks, human ratings, or automated checks, to decide whether it is accurate and safe enough for a given use.
- Model weightsModel weights are the numbers inside a trained neural network that determine how it responds to input. They are what training produces and what a company ships or protects; releasing them publicly is what makes a model 'open weights.'
- Multi-agent systemA multi-agent system is an AI setup in which several agents, each with its own role, instructions, and tools, work together on a task, usually with one agent coordinating the others. It splits complex jobs into pieces the way a team would.
- Multimodal AIMultimodal AI is a model or system that can understand and produce more than one type of content, such as text, images, audio, and video. Modern assistants that can read a screenshot, describe a chart, or listen to speech are multimodal.
N
- Narrow AINarrow AI is artificial intelligence designed and trained for a specific task or a limited set of tasks, such as filtering spam, recognizing faces, or recommending products. Every AI system in use today is narrow in this sense.
- Natural language processing (NLP)Natural language processing (NLP) is the branch of AI focused on understanding and generating human language: reading, classifying, translating, summarizing, and answering questions in ordinary words. Large language models are its current dominant technology.
- Negative promptingNegative prompting is telling an AI model what to avoid or leave out of its output. Image generators often have a separate 'negative prompt' field for unwanted elements; in chat models it is a plain instruction like 'do not use bullet points.'
- Neural networkA neural network is a computing system made of layers of simple connected units that pass numbers to each other, with connection strengths (weights) adjusted during training until the network produces useful output. It is the building block of modern AI.
- No-codeNo-code refers to tools that let people build apps, automations, and websites through visual interfaces and plain-language instructions instead of writing programming code.
O
- Open-weights modelAn open-weights model is an AI model whose trained parameters (weights) are published for anyone to download, run, and modify, usually under a license that sets limits on commercial use or redistribution.
- Optical character recognition (OCR)Optical character recognition (OCR) is technology that converts text in images, scans, and PDFs into machine-readable, searchable, editable text. Modern AI models extend it to read messy layouts, tables, forms, and handwriting.
- OverfittingOverfitting happens when a machine learning model learns its training data too closely, memorizing quirks and noise instead of general patterns, so it performs well on the examples it saw and poorly on new ones.
P
- ParametersParameters are the adjustable numbers inside an AI model, mainly its weights, that are learned during training. The parameter count, from a few billion to over a trillion, is the standard shorthand for a model's size and rough capability.
- Persona promptingPersona prompting is the technique of telling an AI model to respond as a specific role or character, such as 'You are a senior tax accountant,' to shape the expertise, tone, and assumptions in its answer.
- Personally identifiable information (PII)Personally identifiable information (PII) is any data that can identify a specific person on its own or in combination with other data, such as a name, address, Social Security number, email, phone number, photo, or account ID.
- Pre-trainingPre-training is the first and largest phase of building an AI model, in which it learns general patterns from a massive dataset, such as predicting the next word across trillions of words of text. Later phases then shape it into a useful assistant.
- Predictive analyticsPredictive analytics uses historical data and machine learning models to estimate how likely a future outcome is: which customers will churn, which invoices will be paid late, what a store will sell next month. The output is a probability, not a fact.
- PromptA prompt is the text (and sometimes files or images) you give an AI model to tell it what you want. A good prompt states the role the AI should play, the task, the relevant context, the output format, and any constraints.
- Prompt chainingPrompt chaining is the practice of breaking a complex task into a sequence of smaller prompts, where the output of one step becomes the input to the next, instead of asking an AI model to do everything in a single request.
- Prompt engineeringPrompt engineering is the practice of writing and refining instructions to get reliable, high-quality output from an AI model. It covers giving context, showing examples, specifying format, and iterating on what did not work.
- Prompt injectionPrompt injection is an attack in which instructions hidden inside content an AI system reads, such as a web page, email, or document, hijack the system into doing something its user never asked for.
- Prompt templateA prompt template is a reusable, pre-written prompt with placeholders for the parts that change, such as [CLIENT NAME] or [PASTE THE EMAIL], so you can get consistent results from an AI model without rewriting instructions every time.
- Proprietary modelA proprietary model is an AI model whose weights are kept private by the company that built it. You use it through the company's app or API, and you cannot download, inspect, or self-host it.
R
- Rate limitA rate limit is a cap on how much you can use an AI service in a given period, such as messages per hour on a chat plan or requests and tokens per minute on an API, set to protect the provider's capacity and to tier pricing.
- Reasoning modelA reasoning model is a large language model trained to work through a problem step by step before answering, spending extra computation (often called thinking) to improve accuracy on math, code, logic, and multi-step tasks.
- Recommendation systemA recommendation system is software that predicts what a person is most likely to want next, such as a product, an article, or a video, based on past behavior and the behavior of similar people. It drives feeds and "you might also like" boxes across the web.
- Red teamingRed teaming is the practice of deliberately attacking an AI system, by trying to make it produce harmful, false, or policy-violating output, in order to find weaknesses before real users or bad actors do.
- Reinforcement learningReinforcement learning is a way of training AI by trial and error: the system takes actions, receives rewards or penalties based on the outcomes, and adjusts its behavior to earn more reward over time.
- Reinforcement learning from human feedback (RLHF)RLHF is a training step in which human raters compare model responses, their preferences train a reward model, and the language model is then tuned to produce answers people rate highly. It is how raw text predictors became helpful assistants.
- Responsible AIResponsible AI is an approach to building and using AI that deliberately addresses fairness, transparency, privacy, safety, and accountability, so the technology's benefits do not come at the expense of the people affected by it.
- Retrieval-augmented generation (RAG)Retrieval-augmented generation (RAG) is a technique in which an AI system first searches a set of documents for relevant passages and then gives those passages to a language model to answer from. It lets an assistant use your own, current information.
- Robotic process automation (RPA)Robotic process automation (RPA) is software that mimics a person's clicks and keystrokes to run repetitive, rule-based tasks, such as copying invoice data from email into an accounting system. It follows a fixed script and makes no judgment calls.
S
- Semantic searchSemantic search finds results based on the meaning of a query rather than exact keyword matches, so a search for 'unpaid invoices' also returns documents about 'overdue bills.' It uses embeddings to compare meaning.
- Sentiment analysisSentiment analysis is the use of AI to classify the emotional tone of text, typically as positive, negative, or neutral, and often in finer categories such as frustrated, satisfied, or urgent, across large volumes of reviews, tickets, or messages.
- Shadow AIShadow AI is the use of AI tools inside an organization without the knowledge or approval of IT, security, or leadership: personal ChatGPT accounts, browser extensions, and AI features quietly switched on inside existing software.
- Small language model (SLM)A small language model (SLM) is a language model with far fewer parameters than a frontier model, small enough to run on a laptop, phone, or single server. It trades some capability for speed, low cost, and the option to run privately on your own hardware.
- Speech recognitionSpeech recognition, also called speech-to-text or automatic speech recognition (ASR), is AI that converts spoken audio into written text. It powers dictation, meeting transcription, voicemail transcripts, and voice commands.
- Structured outputStructured output is an AI model's answer returned in a fixed, machine-readable format such as JSON or a table with defined fields, rather than free-form prose, so other software can use the result directly without a person re-typing it.
- SummarizationSummarization is the use of AI to condense a long document, transcript, thread, or dataset into a shorter version that keeps the key points. It is one of the most reliable everyday uses of language models, with known failure modes worth checking.
- Supervised learningSupervised learning trains a machine learning model on examples that already carry the correct answer, so it can predict the answer for new cases. It is behind spam filters, credit scoring, demand forecasts, and most business AI that classifies or predicts.
- Synthetic dataSynthetic data is artificially generated data, often produced by an AI model, that mimics the statistical patterns of real data and is used to train, test, or demonstrate systems without exposing real records.
- System promptA system prompt is a set of instructions given to an AI model before a conversation starts that defines its role, tone, rules, and boundaries. Vendors write one for their products; you can add your own in custom instructions, projects, and custom assistants.
T
- TemperatureTemperature is a setting that controls how random a language model's output is. Low values make the model pick the most likely next word and give consistent, predictable answers; higher values allow more varied and creative responses.
- Text-to-imageText-to-image is generative AI that creates a picture from a written description. Tools such as Midjourney, Adobe Firefly, and the image generator built into ChatGPT turn a prompt like 'a product photo of a blue water bottle on white' into a finished image.
- Text-to-speech (TTS)Text-to-speech (TTS) is AI that converts written text into natural-sounding spoken audio. Modern systems produce voices with realistic pacing and emotion, and some can clone a specific person's voice from a short sample.
- TokenA token is the unit of text a language model reads and writes, usually a word or a piece of a word. Models measure input limits, output limits, and pricing in tokens; as a rough rule, 1,000 tokens is about 750 words of English.
- TokenizationTokenization is the process of splitting text into tokens, the small chunks of characters that a language model actually reads and writes. Common words are often one token; rare words, numbers, and code are broken into several.
- Tool useTool use (also called function calling) is the ability of an AI model to request that an outside tool be run, such as a web search, a calculator, a database query, or a calendar lookup, and then use the result in its answer.
- Training dataTraining data is the collection of examples an AI model learns from, such as web pages, books, code, images, or labeled records. What is in that data, and what is missing, shapes everything the model knows, how it writes, and which mistakes it makes.
- Transfer learningTransfer learning reuses a model trained on one large task as the starting point for a different, usually smaller task, so the new model needs far less data and computing. It is the reason a general-purpose AI model can be adapted to your specific work.
- TransformerThe transformer is the neural network architecture behind nearly all modern language models. Its key idea, called attention, lets the model weigh every word in a passage against every other word, so it can track meaning across long stretches of text.
V
- Vector databaseA vector database stores embeddings, the numeric representations of text, images, or other content, and finds the items closest in meaning to a query quickly. It is the storage layer behind semantic search and most document-aware AI assistants.
- Vibe codingVibe coding is building software by describing what you want to an AI coding assistant in plain language and accepting the code it produces, iterating on results rather than reading or writing the code yourself.
- Voice agentA voice agent is an AI system that holds a spoken, real-time conversation with a person, typically over the phone, and can complete tasks such as booking appointments, answering questions, or qualifying leads without a human on the line.
- Voice cloningVoice cloning is the use of AI to create a synthetic copy of a specific person's voice from a short audio sample, so that new text can be spoken in that voice. It powers useful narration tools and also some of the most convincing scams.
W
- Workflow automationWorkflow automation connects apps and steps so that a trigger, such as a new form response or an incoming email, automatically runs a sequence of actions, increasingly with an AI step that reads, decides, or writes along the way.
- Workplace AI policyA workplace AI policy is a written set of rules that tells employees which AI tools they may use, what data they may put into them, how to verify and disclose AI-assisted work, and who to ask when unsure.
Z
- Zero data retention (ZDR)Zero data retention (ZDR) is an arrangement in which an AI provider does not store your prompts or the model's responses after the request is processed, so nothing is kept for training, debugging, or later review. Mainly offered on API and enterprise plans.
- Zero-shot promptingZero-shot prompting is asking an AI model to do a task with instructions only, without showing it any examples of the output you want. It works well for common tasks the model has seen many times, such as summarizing or translating.