GlossaryA to L
Say it in plain English.
You don't need to memorise any of this. Look a word up when it gets in your way, then carry on.
- AI (artificial intelligence) the big one
- Software that does things we used to think needed a human brain: writing, summarising, recognising pictures, answering questions.In a sentence: "I used AI to draft the email, then tidied it up."
- AI agent
- An AI that doesn't just answer but carries out a multi-step job for you, such as researching five hotels and putting them in a table.Think: an assistant you hand a task to, not a question.
- AI assistant (chatbot)
- The chat window you type into. ChatGPT, Claude, Gemini and Copilot are all AI assistants.Where every beginner should start.
- Automation
- Getting a task to run by itself whenever something happens, like saving every email attachment to a folder.Covered in Time Bending.
- Bias
- When an AI's answers lean one way because of the material it learned from. It can repeat stereotypes or miss viewpoints.Counter-move: ask for other perspectives.
- Context
- The background you give the AI so it understands your situation: who it's for, why you need it, what you've tried.More context nearly always means a better answer.
- Context window
- How much the AI can keep in mind at once in a single chat. In a very long chat it can lose track of what was said early on.Counter-move: start a fresh chat and paste in a short recap.
- Custom assistant
- A version of an AI assistant you set up once with your own instructions and notes, so you don't repeat yourself each time.Covered in Sidekick Summoning.
- Data
- Information. For AI, usually the huge amount of text and images it learned from, or the files you give it.
- Fine-tuning
- Giving an existing AI model extra training on specialist examples. Beginners almost never need it.If someone says you must fine-tune to get started, you needn't.
- Generative AI
- AI that creates new things, such as text, pictures, audio or video, instead of only sorting or labelling.
- Hallucination villain alert
- When an AI states something false as if it were fact: an invented quote, a wrong date, a book that doesn't exist.Counter-move: verify names, numbers and sources. See X-Ray Research.
- Image generator
- A tool that turns a written description into a picture.Covered in Visual Conjuring.
Field noteThe only five you need
Short on time? Learn these five.
Rookie essentials
- Promptwhat you ask
- Contextthe background you give
- Hallucinationa confident mistake
- Modelthe engine version
- Automationtasks that run themselves
Everything else is detail. If a tutorial leans on words that aren't on this page, it probably isn't written for beginners.
GlossaryL to W
- LLM (large language model) jargon beast
- The engine inside a chat assistant. It has read an enormous amount of text and predicts which words should come next.You can ignore the term. "The AI" will do.
- Model
- A particular version of an AI engine. Assistants often let you choose between a faster model and a more capable one.If in doubt, use whichever is selected by default.
- Multimodal
- An AI that handles more than text: it can look at photos, listen to speech or read a PDF.Try it: photograph a letter and ask what it means.
- Parameters jargon beast
- The millions or billions of internal dials a model adjusted while learning. A measure of size, not something you ever touch.
- Prompt
- Whatever you type or say to the AI. A question, an instruction, a pasted document with a request.Covered in Prompt Whispering.
- Prompt engineering
- A grand name for getting good at asking. Clear task, useful background, the format you want.
- RAG (retrieval-augmented generation) jargon beast
- When the AI looks things up in a set of documents before answering, instead of relying on memory alone.In plain English: "it checks your files first".
- System prompt (instructions)
- Standing instructions that tell an assistant how to behave in every chat, such as "always reply in British English".
- Token jargon beast
- A chunk of text, roughly three-quarters of a word. AI tools measure lengths and limits in tokens.You'll only meet it when you hit a limit.
- Training
- The process where a model learns patterns from a vast amount of material before you ever use it.
- Workflow
- A set of steps that turns a starting point into a finished result. In automation tools, a workflow is the chain you build.