UseAIEasily Shop
UseAIEasily

AI glossary

AI glossary: AI terms explained simply

This AI glossary explains 84 artificial intelligence terms in one or two sentences, each with an example and a link to the free lesson or course that teaches it, from tokens and prompting to RAG and AI agents.

AI basics

AI (artificial intelligence)

Also called: artificial intelligence, what is ai

AI (artificial intelligence) is the umbrella term for computer systems that handle tasks that normally need human thinking: understanding text, recognising images, suggesting decisions or creating new content.

Example: An accounting firm uses AI to read the tax number, total and due date from incoming invoices, so staff only have to check the data.

Learn it: Free lesson: Models and tokens · Guide: How to start learning AI · Free AI course

Related terms: Generative AI, LLM (large language model), Inference

Generative AI

Also called: genai, gen ai

Generative AI is AI that creates new content, such as text, images, audio, video or code, based on patterns learned from its training data.

Example: A small online shop owner uses generative AI to draft product descriptions for 40 new items and a newsletter header image.

Learn it: Free lesson: Models and tokens · Free lesson: What it can and can't do · Course: AI Basics & Practical Use (ChatGPT, Claude, Gemini)

Related terms: AI (artificial intelligence), LLM (large language model), Multimodal AI, Hallucination

LLM (large language model)

Also called: large language model, language model

An LLM (large language model) is a neural network trained on huge amounts of text that predicts the next piece of text (token) and builds full answers from those predictions.

Example: When you ask ChatGPT to draft a polite reply to an unhappy customer, an LLM writes it token by token.

Learn it: Free lesson: Models and tokens · Free lesson: What it can and can't do · Free demo lesson: AI Basics

Related terms: Token, Transformer, Parameter, GPT (Generative Pre-trained Transformer)

GPT (Generative Pre-trained Transformer)

Also called: generative pre-trained transformer, what does gpt stand for

GPT (Generative Pre-trained Transformer) is the name of OpenAI's family of language models: generative, pre-trained on large amounts of text, and built on the transformer architecture.

Example: In "ChatGPT", GPT is the model running in the background and "Chat" is the conversational interface you type into.

Learn it: Free lesson: Models and tokens · Guide: ChatGPT vs Claude vs Gemini

Related terms: ChatGPT, Transformer, LLM (large language model)

Transformer

Also called: transformer architecture, attention mechanism

The transformer is the neural network architecture behind today's large language models: its attention mechanism looks at all parts of the text at once and works out which words relate to which.

Example: In "Kate gave Anna the book because she had finished it", attention links "she" to Kate.

Learn it: Free lesson: Models and tokens

Related terms: LLM (large language model), GPT (Generative Pre-trained Transformer), Token, Context window

Context window

Also called: context length, context size

The context window is the maximum amount of text, measured in tokens, that a language model can take in at once: conversation history, uploaded files, instructions and its own reply together.

Example: Paste a 300-page contract plus a long email thread and the start of the chat may fall out of the window, so the model "forgets" your earlier instructions.

Learn it: Free lesson: Models and tokens · Free lesson: Context and examples

Related terms: Token, Context engineering, RAG (Retrieval-Augmented Generation), Hallucination

Training data and knowledge cutoff

Also called: knowledge cutoff, training cutoff, cutoff date

Training data is the collection of text, images and code a model learns from. The knowledge cutoff is the date that data ends: the model knows nothing newer on its own.

Example: Ask about a tax rule that changed last week without web search and the model will confidently describe the old one.

Learn it: Free lesson: What it can and can't do · Free lesson: Hallucination · Free lesson: Data and trust

Related terms: Parameter, Hallucination, RAG (Retrieval-Augmented Generation), Fine-tuning

Hallucination

Also called: ai hallucination, llm hallucination, made-up answer

A hallucination is when an AI model confidently states something false: a source that doesn't exist, an invented number, a fake quote or a wrong legal reference.

Example: You ask for three court rulings on a rent dispute and get neatly formatted case numbers, none of which exist.

Learn it: Free lesson: Hallucination · Guide: Why does AI hallucinate? · Free lesson: Check before you send

Related terms: Training data and knowledge cutoff, RAG (Retrieval-Augmented Generation), Temperature, AI evaluation (evals)

Reasoning model

Also called: thinking model, thinking mode, reasoning llm

A reasoning model is a language model that works through a task in internal steps before answering, making it more accurate on complex logic, maths and coding problems.

Example: Hunting a bug in a pricing spreadsheet, thinking mode takes longer but checks each row and finds the wrong VAT rate.

Learn it: Free lesson: What it can and can't do · Free lesson: Picking the right tool

Related terms: Chain-of-thought, LLM (large language model), Inference, Token

Open-weight model

Also called: open weights, open model, open source llm, local llm

An open-weight model is an AI model whose trained parameters (weights) you can download and run on your own hardware, unlike closed models available only through the vendor's cloud.

Example: A private clinic runs an open-weight model on its own server so patient notes never leave the building.

Learn it: Free lesson: Data and trust · Free lesson: Models and tokens

Related terms: Parameter, LLM (large language model), GDPR (General Data Protection Regulation), Fine-tuning

Diffusion model

Also called: diffusion, Stable Diffusion, denoising model

A diffusion model is a type of image-generating AI that starts from random noise and removes it step by step until an image matching the prompt appears. Many well-known image generators, such as Stable Diffusion, work this way.

Example: An image generator showing blurry blobs that gradually sharpen illustrates the diffusion steps.

Learn it: Free lesson: Models and tokens

Related terms: AI image generation (text-to-image), Generative AI, Transformer

AI tools

ChatGPT

Also called: chat gpt, openai chatgpt

ChatGPT is OpenAI's conversational AI assistant built on large language models: it answers questions, writes and rewrites text, reads files and images, and can search the web.

Example: A small business owner pastes monthly sales figures and asks ChatGPT for a five-point summary for Thursday's meeting.

Learn it: Free lesson: Picking the right tool · Guide: ChatGPT vs Claude vs Gemini · Course: AI Basics & Practical Use (ChatGPT, Claude, Gemini)

Related terms: GPT (Generative Pre-trained Transformer), Claude, Gemini (Google Gemini), Prompt

Claude

Also called: claude ai, anthropic claude, claude.ai

Claude is Anthropic's AI assistant and model family, strong at long documents, careful writing and coding, available as a chat app, through an API and in developer tools.

Example: A legal assistant uploads a 60-page lease and asks Claude to tabulate the termination clauses and deadlines with page numbers.

Learn it: Free lesson: Picking the right tool · Guide: ChatGPT vs Claude vs Gemini · Course: AI Basics & Practical Use (ChatGPT, Claude, Gemini)

Related terms: ChatGPT, Gemini (Google Gemini), MCP (Model Context Protocol), Context window

Gemini (Google Gemini)

Also called: google gemini, google bard, bard

Gemini is Google's AI model family and assistant (formerly Bard), available as its own app and built into Google services such as Gmail, Docs and Sheets.

Example: An office manager asks Gemini in Gmail to summarise a 25-message supplier thread and draft a reply.

Learn it: Free lesson: Picking the right tool · Guide: ChatGPT vs Claude vs Gemini · Course: AI Basics & Practical Use (ChatGPT, Claude, Gemini)

Related terms: ChatGPT, Claude, Multimodal AI

Microsoft Copilot

Also called: copilot, microsoft 365 copilot, github copilot

Microsoft Copilot is Microsoft's AI assistant built into Windows, the Edge browser and Microsoft 365 apps (Word, Excel, Outlook, Teams), so it works where your company documents live.

Example: After a Teams meeting, Copilot lists who agreed to what, and the project lead turns it into minutes in Word.

Learn it: Free lesson: Picking the right tool · Free lesson: Summaries and drafts

Related terms: ChatGPT, Gemini (Google Gemini), Cursor (AI code editor)

Cursor (AI code editor)

Also called: cursor ai, cursor editor, ai code editor

Cursor is an AI-first code editor built on VS Code: it understands your whole project, edits code across files from plain-language requests and can run commands in agent mode.

Example: You ask Cursor to add a coupon field to the checkout page and review the changes it proposes before accepting them.

Learn it: Course: AI-Powered Web & App Development (Vibe Coding) · Guide: What is vibe coding? · Free lesson: What vibe coding is

Related terms: Vibe coding, Microsoft Copilot, Claude, AI agent

n8n (workflow automation)

Also called: n8n automation, n8n workflow, n8n self-hosted

n8n is a visual workflow automation tool where you build processes from nodes: it connects apps, runs AI steps and agents, and its source-available code lets you self-host it.

Example: An n8n workflow watches Gmail for invoices, has AI extract the supplier and amount, logs them in Google Sheets and pings the accountant.

Learn it: Guide: Automating with n8n · Free demo lesson: AI Automation · Free lesson: Workflows, triggers, actions

Related terms: Workflow, Make (Make.com, formerly Integromat), AI automation, Webhook

Make (Make.com, formerly Integromat)

Also called: make.com, integromat, make scenario

Make (Make.com, formerly Integromat) is a cloud-based no-code automation platform where you connect apps in visual scenarios and can add AI steps.

Example: A Make scenario creates an invoice for every new shop order, saves the PDF to Google Drive and emails the customer.

Learn it: Course: AI Automation & Workflow Building (N8N, Make.com) · Free lesson: Workflows, triggers, actions · AI automation course

Related terms: n8n (workflow automation), Workflow, Trigger, AI automation

DALL-E

Also called: DALL·E, DALLE, OpenAI image generation, ChatGPT image generation

DALL-E is the name of OpenAI's text-to-image model family. ChatGPT generates images with OpenAI's image models, so in everyday speech “DALL-E” often just means image generation in ChatGPT.

Example: A shop owner asks ChatGPT for a newsletter header illustration, then refines it: fewer elements, shades of green.

Learn it: Course: AI Basics & Practical Use (ChatGPT, Claude, Gemini)

Related terms: AI image generation (text-to-image), Midjourney, ChatGPT, Multimodal AI

AI image generation (text-to-image)

Also called: text-to-image, AI image generator, AI art

AI image generation means a model creates a new image from a text description (prompt) or edits an existing one. Well-known tools include ChatGPT, Gemini, Midjourney and Adobe Firefly.

Example: A marketer generates three colour variants of one ad scene and A/B tests them.

Learn it: Course: AI Basics & Practical Use (ChatGPT, Claude, Gemini) · Free lesson: Anatomy of a good prompt

Related terms: Diffusion model, Midjourney, DALL-E, AI content labelling (watermarks, C2PA / Content Credentials)

Chatbot (AI chatbot)

Also called: AI chatbot, virtual assistant, customer service bot

A chatbot is a conversational program that answers users in text or voice. Modern AI chatbots run on a language model and, ideally, answer from the company's own knowledge base rather than made-up information.

Example: An online store's chatbot answers delivery and returns questions from the policy and hands complaints to a person.

Learn it: Free lesson: RAG in plain words · Course: AI Agents & Automated Systems (No-code, Low-code, Code)

Related terms: Knowledge base, RAG (Retrieval-Augmented Generation), Human-in-the-loop, Transparency obligations (AI Act Article 50)

Custom GPT

Also called: GPTs, ChatGPT custom GPT, build a GPT

A custom GPT is an assistant you configure inside ChatGPT for one job, with its own instructions, uploaded files and optional tools, so you don't re-explain the task in every chat.

Example: A real estate agency builds a “listing writer” GPT with its style guide and five sample listings.

Learn it: Free lesson: Role and audience · Course: AI Basics & Practical Use (ChatGPT, Claude, Gemini)

Related terms: ChatGPT, System prompt, Knowledge base

NotebookLM

Also called: Google NotebookLM, Notebook LM, Gemini Notebook

NotebookLM is Google's AI notebook that answers from the sources you upload (PDFs, Google Docs, websites) and shows which passage each answer came from, which makes it less prone to making things up. Since July 2026 it is called Gemini Notebook; it is the same product.

Example: An HR manager uploads the staff handbook and asks about leave rules, with the cited paragraph shown.

Learn it: Free lesson: Summaries and drafts

Related terms: RAG (Retrieval-Augmented Generation), Gemini (Google Gemini), Hallucination, Knowledge base

Deep Research

Also called: deep research mode, AI research mode, research agent

Deep Research is a mode in several AI assistants (such as ChatGPT and Gemini) where the model searches and reads many sources on its own for several minutes, then writes a report with citations.

Example: A business owner asks for a market overview and verifies each cited figure at its source.

Learn it: Guide: AI tools compared · Free lesson: Check before you send

Related terms: Perplexity, AI agent, Hallucination, Reasoning model

Prompting

Prompt

Also called: prompts, prompting, ai prompt

A prompt is the instruction or question you give an AI model; answer quality depends largely on how clearly it states the task, context and expected format.

Example: Instead of "Write an email", try "Write a short, friendly apology to a customer whose parcel is 3 days late, offer a 10% coupon, max 80 words."

Learn it: Guide: How to write a good prompt · Free lesson: Anatomy of a good prompt · Practise free in the sandbox

Related terms: Prompt engineering, System prompt, Few-shot prompting, Context window

Prompt engineering

Also called: prompt design, prompt writing

Prompt engineering is the deliberate design, testing and refinement of prompts so an AI model answers reliably, at the quality and in the format you need.

Example: A support team tests its reply prompt on 20 real customer emails and tightens the rules until all 20 answers are acceptable.

Learn it: Free lesson: Anatomy of a good prompt · Prompt engineering course · Free demo lesson: AI Basics

Related terms: Prompt, Context engineering, Few-shot prompting, Chain-of-thought

System prompt

Also called: system message, custom instructions

A system prompt is the base instruction a model receives before the conversation and that stays in force throughout: it sets the role, tone, rules and what the model must not do.

Example: A shop chatbot's system prompt: "You are Garden Shop support. Only discuss products, shipping and returns; quote prices only from the given list."

Learn it: Free lesson: Role and audience · Free lesson: Prompt injection

Related terms: Prompt, Context engineering, Prompt injection, Guardrail

Few-shot prompting

Also called: few shot, few-shot, one-shot, examples in prompt

Few-shot prompting means including a few (typically 2-5) worked examples in the prompt so the model sees what output you expect for what input and follows the pattern.

Example: Paste three existing product descriptions from your shop and ask for a new one in the same style and length.

Learn it: Free lesson: Few-shot, done right · Free lesson: Context and examples

Related terms: Zero-shot prompting, Prompt, Structured output, Fine-tuning

Chain-of-thought

Also called: chain of thought, cot, step by step, think step by step

Chain-of-thought is a prompting technique where you ask the model to work through the solution step by step before the final answer, which reduces errors on maths and multi-step tasks.

Example: "Work out the 3-month cost for a team of 4 at a monthly per-seat price with a 10% quarterly discount. Show the steps, then the total."

Learn it: Free lesson: Structured output and self-checks · Free lesson: Iteration and evaluation

Related terms: Reasoning model, Prompt engineering, Zero-shot prompting

Context engineering

Also called: context design, context management

Context engineering is deciding exactly what information goes into the model's context window at each step: instructions, document snippets, history, tool results, and what to leave out.

Example: A support agent gets only the customer's last 3 orders, the relevant returns rule and the system prompt, not the whole 200-page handbook.

Learn it: Free lesson: Context and examples · Free lesson: RAG in plain words · Free lesson: Agents, tools and MCP

Related terms: Prompt engineering, Context window, RAG (Retrieval-Augmented Generation), System prompt

Structured output

Also called: json output, json mode, json schema output

Structured output is when a model returns data in a predefined shape, usually JSON with fixed field names, instead of free text, so a program or workflow can process it reliably.

Example: An automation asks for {"supplier", "tax_id", "total", "due_date"} from each invoice so the result goes straight into a spreadsheet.

Learn it: Free lesson: Structured output and self-checks · Free lesson: AI inside a workflow

Related terms: Function calling (tool use), Few-shot prompting, Temperature, Workflow

Brand voice

Also called: tone of voice, style guide, brand tone

Brand voice is the consistent way a company writes and speaks: how formal, how direct, which words it uses and avoids. For AI writing you pass it in the prompt or an assistant's instructions, otherwise the text sounds generic.

Example: A craft brewery describes its voice as casual, funny and short, never “premium”, and adds three past posts as examples.

Learn it: Free lesson: Role and audience · Free lesson: Context and examples

Related terms: System prompt, Few-shot prompting, Prompt, Custom GPT

Building with AI

API

Also called: application programming interface, AI API, LLM API

An API (application programming interface) is a defined way for two programs to talk: one sends a request, the other returns a response. With AI, it is how your own code or automation reaches a model without a chat window.

Example: An online store sends each new product's specs to a language model via API and gets back a ready-to-publish product description.

Learn it: Free lesson: APIs and keys · Course: AI Agents & Automated Systems (No-code, Low-code, Code)

Related terms: API key, Function calling (tool use), Webhook, LLM (large language model)

API key

Also called: secret key, API token, OpenAI API key

An API key is a long secret string that identifies you to a provider when your code calls a model through its API. Treat it like a password to your company card: whoever has it spends on your account.

Example: A team stores its model API key in n8n's credential store rather than in a shared doc, and sets a monthly spending limit with the provider.

Learn it: Free lesson: APIs and keys · Free lesson: Ship it safely

Related terms: API, AI security, RBAC (role-based access control)

Embedding

Also called: embeddings, vector embedding, text embedding

An embedding is a text's meaning turned into a list of numbers (a vector), where texts with similar meaning end up close together. It lets software search by meaning, not just matching words.

Example: In a help center, the question "How do I get my money back?" finds the article "Returns and refunds" even though they share no words.

Learn it: Guide: What is RAG? · Free lesson: RAG in plain words

Related terms: Vector database, RAG (Retrieval-Augmented Generation), Chunking

Chunking

Also called: chunks, text splitting, document splitting

Chunking is splitting long documents into smaller, self-contained pieces before embedding them. Good chunking decides whether a RAG system retrieves the exact paragraph or half a sentence.

Example: A store's terms and conditions are split by section, with each chunk keeping its heading, so a shipping question returns the whole shipping clause.

Learn it: Free lesson: RAG in plain words · Guide: What is RAG?

Related terms: RAG (Retrieval-Augmented Generation), Embedding, Context window

RAG (Retrieval-Augmented Generation)

Also called: retrieval augmented generation, RAG pipeline, chat with your documents

RAG (Retrieval-Augmented Generation) is a method where the system first retrieves relevant passages from your own documents, then asks the language model to answer using them. The model works from your data, not only from what it saw in training.

Example: An internal assistant answers a policy question from the company handbook and shows which document it used.

Learn it: Guide: What is RAG? · Free lesson: RAG in plain words · Course: AI Agents & Automated Systems (No-code, Low-code, Code)

Related terms: Embedding, Vector database, Chunking, Hallucination

Fine-tuning

Also called: fine tuning, finetuning, model customization

Fine-tuning is further training an already trained model on your own examples so it reliably follows a specific style, format or task. It mainly adjusts behavior rather than adding new knowledge.

Example: An insurer fine-tunes a model on thousands of past claim replies so every new letter matches the company's tone and structure.

Learn it: Free lesson: RAG in plain words · Guide: What is RAG?

Related terms: RAG (Retrieval-Augmented Generation), Training data and knowledge cutoff, Open-weight model, Few-shot prompting

Function calling (tool use)

Also called: tool use, tool calling, function call

Function calling (or tool use) lets a language model signal that one of your defined tools should be called, and with what arguments. Your code runs the call and hands the result back to the model.

Example: A customer asks a store chatbot where order 10482 is; the model calls an order_status tool and writes a reply from the tracking data.

Learn it: Free lesson: Agents, tools and MCP · Free demo lesson: AI Agents

Related terms: AI agent, MCP (Model Context Protocol), API, Structured output

AI agent

Also called: agent, autonomous agent, agentic AI, LLM agent

An AI agent is a system where a language model gets a goal and decides step by step which tools to use until the task is done. A chatbot answers; an agent does the work.

Example: An agent reads supplier quotes from Gmail each morning, logs prices in Google Sheets and drafts a reply that a buyer approves before it is sent.

Learn it: Guide: What is an AI agent? · Free demo lesson: AI Agents · Free lesson: Agents, tools and MCP

Related terms: Function calling (tool use), Multi-agent system, MCP (Model Context Protocol), Human-in-the-loop

Multi-agent system

Also called: multi agent, multiagent, agent team, agent orchestration

A multi-agent system is several cooperating AI agents, each with a narrower role such as researcher, writer or reviewer. A coordinating agent or a fixed workflow hands out the work.

Example: One agent gathers product data, a second writes the newsletter, a third checks prices and tone before a human sees it.

Learn it: Free demo lesson: AI Agents · Guide: What is an AI agent? · Free lesson: Agents, tools and MCP

Related terms: AI agent, Workflow, AI evaluation (evals)

MCP (Model Context Protocol)

Also called: model context protocol, MCP server, MCP client

MCP (Model Context Protocol) is an open standard that lets AI applications connect to external tools and data sources in a uniform way. Think of it as a universal connector: one MCP server works with many AI clients.

Example: A team writes an MCP server for its ticketing system and can then query open bugs from both Claude and Cursor.

Learn it: Free lesson: Agents, tools and MCP · Course: AI Agents & Automated Systems (No-code, Low-code, Code)

Related terms: Function calling (tool use), AI agent, Claude, API

Vibe coding

Also called: vibecoding, AI coding, coding with AI

Vibe coding is building software by describing what you want in plain language while the AI writes the code; you test, give feedback and steer. The focus is on the goal and the result, not typing lines by hand.

Example: A salon owner builds a simple booking page in an afternoon by describing services and opening hours to Cursor and asking for fixes until mobile looks right.

Learn it: Guide: What is vibe coding? · Free lesson: What vibe coding is · Vibe coding course

Related terms: Cursor (AI code editor), Prompt, AI security

AI evaluation (evals)

Also called: evals, LLM evaluation, AI testing, model evaluation

AI evaluation (evals) is the practice of regularly and measurably checking an AI system's quality by scoring its answers on a prepared set of test cases. Without it, "it seems better" is just a feeling.

Example: Before launching a support bot, a team runs 50 real, anonymized customer questions after every prompt change and only ships if accuracy holds.

Learn it: Free lesson: Iteration and evaluation · Free lesson: Check before you send

Related terms: Hallucination, Guardrail, Human-in-the-loop, Prompt engineering

Speech-to-text

Also called: STT, speech recognition, transcription, voice to text

Speech-to-text turns spoken audio into written text. Modern AI models transcribe many languages well, so meetings, calls and voice notes become searchable and summarisable.

Example: A support team transcribes recorded calls and an LLM extracts the most common complaints for a weekly report.

Learn it: Free lesson: AI inside a workflow

Related terms: Text-to-speech, Multimodal AI, Chatbot (AI chatbot), PII and PII redaction

Text-to-speech

Also called: TTS, speech synthesis, read aloud, AI voice

Text-to-speech turns written text into natural-sounding speech. It is used in phone assistants, accessibility and learning materials.

Example: The lessons on the UseAIEasily free learning path can be read aloud, so you can follow them by listening.

Learn it: Free lesson: Models and tokens

Related terms: Speech-to-text, Multimodal AI, Deepfake

Knowledge base

Also called: KB, help center, internal wiki, FAQ database

A knowledge base is a company's organised, up-to-date collection of information (FAQs, policies, product and process docs) that people and AI systems answer from. An AI chatbot or RAG system is only as good as the knowledge base behind it.

Example: An online store keeps shipping, warranty and returns rules in short articles that its chatbot also answers from.

Learn it: Free lesson: RAG in plain words · Course: AI Agents & Automated Systems (No-code, Low-code, Code)

Related terms: RAG (Retrieval-Augmented Generation), Chatbot (AI chatbot), Embedding, Vector database

Automation

AI automation

Also called: AI workflow automation, intelligent automation, automation with AI

AI automation is automating a repetitive process where some steps, such as reading, classifying or writing, are done by an AI model. Classic automation follows rules; AI automation also handles fuzzy, text-heavy steps.

Example: An n8n workflow sorts incoming support emails, finds the right invoice and drafts a reply that an agent sends with one click.

Learn it: Free lesson: AI inside a workflow · Free demo lesson: AI Automation · AI automation course

Related terms: Workflow, n8n (workflow automation), Make (Make.com, formerly Integromat), Human-in-the-loop

Workflow

Also called: automation workflow, process, n8n workflow, Make scenario

A workflow is a chain of steps started by an event (trigger) that ends in a result. In tools like n8n or Make you build it visually from nodes.

Example: A new order creates an invoice, logs a row in Google Sheets and sends a thank-you email, with no manual steps.

Learn it: Free lesson: Workflows, triggers, actions · Guide: Automating with n8n

Related terms: Trigger, AI automation, n8n (workflow automation), Make (Make.com, formerly Integromat)

Trigger

Also called: triggers, event trigger, scheduled trigger, cron

A trigger is the event that starts an automated workflow: an incoming email, a new spreadsheet row, a webhook call or a scheduled time. Without a trigger, a workflow never runs.

Example: Every weekday at 8 am a schedule trigger starts a workflow that collects yesterday's invoices from Gmail and files them by client.

Learn it: Free lesson: Workflows, triggers, actions · Guide: Automating with n8n

Related terms: Workflow, Webhook, AI automation

Webhook

Also called: webhooks, callback URL, webhook trigger

A webhook is a URL that another system automatically calls when something happens on its side. Like a doorbell: instead of checking every minute for new orders, the system tells you.

Example: A store platform sends a webhook to n8n on every successful payment, which instantly issues the invoice and notifies the warehouse.

Learn it: Guide: Automating with n8n · Free lesson: Workflows, triggers, actions · Course: AI Automation & Workflow Building (N8N, Make.com)

Related terms: Trigger, API, Workflow

Human-in-the-loop

Also called: human in the loop, HITL, human approval, human review

Human-in-the-loop means a person checks or approves an AI process's output at defined points before it goes live. The AI prepares, the human decides.

Example: The AI drafts a reply to an upset customer and suggests a discount code, but nothing is sent until a support agent approves it.

Learn it: Free lesson: AI inside a workflow · Guide: What is an AI agent?

Related terms: AI agent, AI automation, Guardrail, Hallucination

Safety

Prompt injection

Also called: indirect prompt injection, prompt injection attack

Prompt injection is an attack where someone hides instructions in text the model processes, so the AI does what the attacker wants instead of its original task. It is especially risky for agents that read emails, web pages or files and can use tools.

Example: A CV contains invisible white text telling the AI screener to rank this candidate first.

Learn it: Guide: What is prompt injection? · Free lesson: Prompt injection

Related terms: Jailbreak, Guardrail, AI security, AI agent

Jailbreak

Also called: AI jailbreak, jailbreaking, ChatGPT jailbreak

A jailbreak is a prompt or trick a user employs to make an AI model break its own safety rules or the limits set in its system prompt. Unlike prompt injection, which hides third-party instructions, a jailbreak comes from the person chatting.

Example: A user role-plays with a store chatbot until it invents a 90% discount code.

Learn it: Free lesson: Prompt injection · Guide: What is prompt injection?

Related terms: Prompt injection, Guardrail, System prompt

Guardrail

Also called: guardrails, AI guardrails, safety filter

A guardrail is a check or rule around an AI system's input and output that stops or fixes anything that shouldn't pass. You don't trust the model alone with safety; you build a filter around it.

Example: A bank's support bot masks pasted card numbers before they reach the model and blocks replies that sound like investment advice.

Learn it: Free lesson: Prompt injection · Free lesson: Ship it safely

Related terms: Prompt injection, Jailbreak, PII and PII redaction, Human-in-the-loop

PII and PII redaction

Also called: PII, personally identifiable information, personal data, data anonymization, redaction

PII (personally identifiable information) is any data that can identify a person: name, email, phone number, address, tax ID, bank account. PII redaction means removing or masking it before text reaches an AI model.

Example: Before summarizing 200 complaint emails with AI, a script replaces every name with [CUSTOMER] and every phone number with [PHONE].

Learn it: Free lesson: Data and trust · Free lesson: Ship it safely

Related terms: GDPR (General Data Protection Regulation), Guardrail, AI security

AI security

Also called: LLM security, AI cybersecurity, securing AI systems

AI security is the set of practices that protect an AI system from misuse, data leaks and manipulation. Classic IT security (passwords, permissions, logging) still applies, plus new threats such as prompt injection.

Example: An internal assistant only sees documents the employee may already access, keys are stored encrypted, and every request is logged.

Learn it: Free lesson: Ship it safely · Guide: What is prompt injection? · Free lesson: Data and trust

Related terms: Prompt injection, API key, RBAC (role-based access control), Guardrail

Deepfake

Also called: deep fake, voice cloning, AI fraud, synthetic media

A deepfake is a realistic fake image, audio clip or video made with AI, showing someone saying or doing something they never did. Convincing voice clones can be made from just a short audio sample.

Example: A finance clerk gets a call in the CEO's voice asking for an urgent transfer and verifies it by calling the CEO back on a known number.

Learn it: Free lesson: Check before you send · Free lesson: Data and trust

Related terms: Generative AI, EU AI Act, AI security

Shadow AI

Also called: unsanctioned AI, unapproved AI tools, BYOAI

Shadow AI is when employees use AI tools for work without the company's knowledge or rules, for example pasting client data into a personal chatbot account. The risk is the uncontrolled data flow, not AI itself.

Example: A firm finds staff pasting payroll data into a free chatbot and replaces the ban idea with a business plan, a short policy and training.

Learn it: Free lesson: Data and trust

Related terms: AI policy (company AI use policy), PII and PII redaction, GDPR (General Data Protection Regulation), AI security

AI content labelling (watermarks, C2PA / Content Credentials)

Also called: AI label, watermark, C2PA, Content Credentials, AI-generated label

AI content labelling makes it visible or machine-readable that an image, video, audio or text was created or altered by AI. A common technical approach is the open C2PA standard (Content Credentials), which writes provenance data into the file's metadata.

Example: A travel agency marks an AI-generated beach image in its ad as “AI illustration” so nobody takes it for a real hotel photo.

Learn it: Free lesson: Data and trust

Related terms: Deepfake, Transparency obligations (AI Act Article 50), AI image generation (text-to-image), EU AI Act

Law and compliance

GDPR (General Data Protection Regulation)

Also called: General Data Protection Regulation, data protection, GDPR and AI

The GDPR (General Data Protection Regulation) is the EU's personal data protection law, applicable since 25 May 2018 to any organization handling data of people in the EU. It applies to AI too: if you feed personal data to a model, that processing must comply.

Example: Before using AI to summarize client emails, a firm checks whether the provider signs a data processing agreement, where data is stored and whether inputs are used for training.

Learn it: Free lesson: Data and trust

Related terms: PII and PII redaction, EU AI Act, AI security

EU AI Act

Also called: AI Act, EU AI regulation, European AI Act

The EU AI Act is the European Union's regulation on artificial intelligence, governing AI systems by risk level: some uses are banned, high-risk ones face strict requirements, most others mainly need transparency. It entered into force on 1 August 2024, with obligations applying in stages.

Example: A recruiter using AI to screen CVs falls into the high-risk category and needs documentation, human oversight and candidate information; an internal email summarizer carries far fewer duties.

Learn it: Free lesson: Data and trust · Free lesson: Agents and automation

Related terms: GDPR (General Data Protection Regulation), Deepfake, AI security, DORA (Digital Operational Resilience Act)

RBAC (role-based access control)

Also called: role-based access control, role based access, permissions, access control

RBAC (role-based access control) assigns permissions to roles such as support, finance or management instead of to individuals, and everyone gets access according to their role. In AI systems it decides which data can appear in a model's answers for whom.

Example: An internal RAG assistant only retrieves financial reports for finance and management roles, so a support agent never sees them.

Learn it: Free lesson: Ship it safely · Free lesson: Data and trust

Related terms: AI security, RAG (Retrieval-Augmented Generation), API key

DORA (Digital Operational Resilience Act)

Also called: Digital Operational Resilience Act, DORA regulation, DORA compliance

DORA (Digital Operational Resilience Act) is an EU regulation requiring financial entities such as banks, insurers and investment firms to keep their IT systems resilient against disruptions and attacks. It applies from 17 January 2025.

Example: An insurer adopting an AI claims tool from an outside vendor must register the vendor as an ICT third-party provider, assess its risks and plan for an outage.

Learn it: Free lesson: Data and trust

Related terms: EU AI Act, GDPR (General Data Protection Regulation), AI security

AI literacy

Also called: AI Act Article 4, AI skills, AI competence

AI literacy is the knowledge needed to use AI systems in an informed way, aware of their risks. Under Article 4 of the EU AI Act, providers and deployers of AI systems must take measures to ensure sufficient AI literacy of their staff; this applies since 2 February 2025.

Example: A 30-person logistics firm runs a short internal training on what its AI tools are for and what not to paste in, and records who attended.

Learn it: Free lesson: Data and trust · Free lesson: What it can and can't do

Related terms: EU AI Act, AI policy (company AI use policy), Shadow AI

High-risk AI system

Also called: high-risk AI, AI Act Annex III, high risk AI system

Under the EU AI Act, high-risk AI systems are those that can significantly affect people's health, safety or fundamental rights, for example in recruitment, education or credit scoring (Annex III). They face strict requirements such as risk management, documentation, human oversight and logging.

Example: HR software that ranks applicants is high-risk; an internal email summariser is not.

Learn it: Free lesson: Data and trust

Related terms: EU AI Act, Human-in-the-loop, Data protection impact assessment (DPIA), Transparency obligations (AI Act Article 50)

Transparency obligations (AI Act Article 50)

Also called: AI Act Article 50, AI disclosure, chatbot disclosure

The transparency obligations in Article 50 of the EU AI Act require that people know when they interact with an AI system such as a chatbot, and that artificially generated or manipulated realistic content (deepfakes) is labelled.

Example: An insurer's chat window opens with: “Hi, I'm the AI assistant. Type ‘agent’ to talk to a person.”

Related terms: EU AI Act, AI content labelling (watermarks, C2PA / Content Credentials), Chatbot (AI chatbot), Deepfake

Data processing agreement (DPA)

Also called: DPA, GDPR Article 28, processor agreement

A data processing agreement (DPA) is the contract under GDPR Article 28 between a company (controller) and a service provider (processor) that sets what the provider may do with the personal data it handles. If client data goes into an AI service, you typically need one.

Example: A dental clinic checks that the AI provider's business plan offers a DPA before summarising patient emails.

Learn it: Free lesson: Data and trust

Related terms: GDPR (General Data Protection Regulation), Data protection impact assessment (DPIA), PII and PII redaction, NAIH (Hungarian data protection authority)

Data protection impact assessment (DPIA)

Also called: DPIA, GDPR Article 35, impact assessment

A data protection impact assessment (DPIA) is the prior analysis required by GDPR Article 35 when processing is likely to pose a high risk to people's rights, for example with new technology or large amounts of sensitive data. It is often the first legal step of an AI rollout.

Example: A private clinic documents data flows, risks and safeguards before adopting an AI report summariser.

Learn it: Free lesson: Data and trust

Related terms: GDPR (General Data Protection Regulation), Data processing agreement (DPA), High-risk AI system, NAIH (Hungarian data protection authority)

NAIH (Hungarian data protection authority)

Also called: NAIH, Hungarian DPA, Hungarian National Authority for Data Protection and Freedom of Information

NAIH (Nemzeti Adatvédelmi és Információszabadság Hatóság) is Hungary's data protection supervisory authority: it enforces the GDPR, handles complaints and can impose fines. It is the competent authority when a Hungarian company processes personal data with AI.

Example: A customer complains that a Hungarian shop's chatbot stored their chat without consent; the complaint goes to NAIH.

Learn it: Free lesson: Data and trust

Related terms: GDPR (General Data Protection Regulation), Data protection impact assessment (DPIA), Data processing agreement (DPA)

AI policy (company AI use policy)

Also called: AI acceptable use policy, AI guidelines, company AI rules

An AI policy is a short internal document stating which AI tools staff may use, which data may and may not go into them, who checks the output and how AI-made content is labelled.

Example: A 15-person agency's one-page policy: company ChatGPT and Claude accounts only, client data anonymised, a human reviews everything sent out.

Learn it: Free lesson: Data and trust

Related terms: Shadow AI, AI literacy, GDPR (General Data Protection Regulation), PII and PII redaction

AI pilot (proof of concept)

Also called: pilot project, proof of concept, PoC, AI trial

An AI pilot is a small, time-boxed trial in which a company tests on one well-defined task with a few users whether an AI solution delivers measurable results before rolling it out.

Example: A freight company tests AI-drafted replies with two agents for four weeks and agrees upfront to stop below a 20% time saving.

Learn it: Free lesson: When to automate

Related terms: AI ROI (return on investment), AI evaluation (evals), AI automation, AI policy (company AI use policy)

AI ROI (return on investment)

Also called: AI ROI, return on investment, AI business case

AI ROI shows how much the gains of an AI rollout (time saved, fewer errors, more revenue) exceed its total cost (subscriptions, development, training, maintenance and time spent checking output).

Example: If an employee saves 5 hours a week, multiply by their hourly cost, then subtract tool fees and review time.

Learn it: Free lesson: When to automate

Related terms: AI pilot (proof of concept), AI automation, AI evaluation (evals)

GEO (generative engine optimization)

Also called: generative engine optimization, AEO, answer engine optimization, AI SEO, LLM SEO

GEO (generative engine optimization) means shaping content so AI answer engines such as ChatGPT search, Perplexity and Google's AI Overviews find, understand and cite it. A related term is AEO (answer engine optimization).

Example: An accounting firm opens each service page with a one-sentence answer, ends with FAQs and adds structured data.

Learn it: Free lesson: Picking the right tool

Related terms: Perplexity, ChatGPT, Structured output, Deep Research

Guides

UseAIEasily at a glance

What is it?
Online AI courses and a free AI learning path for beginners and for work.
Free
AI learning path: 12 units, 36 lessons of about 3 minutes. No sign-up needed for any lesson. Plus the sandbox and 8 demo lessons.
Paid courses
6 hands-on courses, 90 written lessons in total (in Hungarian), €69.00 to €99.00, one-time price, lifetime access.
Premium
9 990 Ft/month (or 99 990 Ft/year): every course and template while subscribed. Billed in HUF.
Languages
Hungarian (primary), English, German.
Who makes it?
DField Solutions, an AI engineering studio in Budapest that builds production AI systems for companies.

DField SolutionsAbout →