AI, from A to Z.
A word. A real-life example. A little “oh, now I get it.”
Choose a letter and take one small step. No coding, no need to memorise everything.
Where shall we start?
Follow A → B → C, or jump to any letter. Some terms are optional deeper dives.
Search a term, its full name or an everyday phrase.
Just starting? These six words are enough for a first try.
A is for
AI.
AI — artificial intelligenceA broad name for computer systems that do tasks such as recognising patterns, making predictions or generating content.Example & what to remember
Start here
Picture this
An email spam filter and an assistant that drafts a reply both use AI, but they do different jobs.
What to remember
AI is a category, not one app. Being useful at one task does not mean a system is good at every task.
AgentAn AI-based system that can choose and take steps toward a goal, often by using tools.Example & what to remember
Useful next
Picture this
An assistant searches approved documents, compares the information and prepares a draft response. Sending it would be a separate action.
What to remember
Check what it can access and what it can do. A helpful answer and permission to take action are different things.
AlgorithmA set of steps a computer follows to solve a problem or complete a task.Example & what to remember
Deeper dive · Optional for your first steps
Picture this
Sorting a list of names alphabetically uses an algorithm. It does not need to use AI.
What to remember
Not every algorithm is AI. Some follow fixed rules; others help a system learn patterns from data.
API — application programming interfaceA defined way for one piece of software to request information or actions from another.Example & what to remember
Deeper dive · Optional for your first steps
Picture this
A booking app asks a calendar service to create an event, instead of a person copying the details manually.
What to remember
You do not need an API to use an ordinary AI chat app. API access may have separate permissions and usage charges.
AutomationMaking a defined task or sequence run with less manual work.Example & what to remember
Useful next
Picture this
A form submission automatically adds a row to a spreadsheet. No AI is required for that rule.
What to remember
Try a process manually and check it before automating it. Automation can repeat mistakes as well as useful steps.
One little check
Which of these can use AI?
B is for
Bias.
BiasA systematic skew in data or results. In everyday AI use, it can mean patterns that unfairly favour or disadvantage people or groups.Example & what to remember
Useful next
Picture this
A generated list of leaders includes only one kind of background, even though your request was broad.
What to remember
Ask whose experiences are missing. A confident-looking result can still reflect a narrow or unfair pattern.
One little check
An AI list includes only one kind of background. What should you do?
C is for
Context.
ContextThe background information available to help the tool interpret your request.Example & what to remember
Start here
Picture this
“Write an invitation” is vague. Adding “for first-time attendees, under 80 words, using these confirmed details” gives useful context.
What to remember
Include what changes the answer: audience, source facts, constraints and what is unknown. Do not assume the tool knows your situation.
ChatbotSoftware you interact with through a conversation. It may use fixed rules, an AI model or a mixture of both.Example & what to remember
Useful next
Picture this
A website helper asks what you need and replies in a chat window.
What to remember
A chat interface does not tell you how capable the system is or whether a human is reviewing its answers.
Context windowThe amount of information a model can work with at one time, usually measured in tokens.Example & what to remember
Deeper dive · Optional for your first steps
Picture this
A very long conversation or large document may exceed what the model can use in a single request.
What to remember
This is not the same as saved memory. Restate important instructions and select relevant material when a task gets long.
One little check
Which brief gives more useful context?
D is for
Data.
DataInformation a system can use, such as words, numbers, images or sound.Example & what to remember
Useful next
Picture this
Your sales totals, meeting notes and product photos are all data, even though they look different.
What to remember
Check whether the information is relevant, accurate and permitted to be shared with the tool.
Deep learningA type of machine learning that uses neural networks with many layers to learn patterns.Example & what to remember
Deeper dive · Optional for your first steps
Picture this
A system learns patterns in labelled photos that help it recognise objects in new pictures.
What to remember
“Deep” describes the system's structure. It does not mean the system has deep understanding like a person.
One little check
Which counts as data?
E is for
Evaluation.
EvaluationChecking how well a system performs against clear criteria and examples.Example & what to remember
Useful next
Picture this
You test an action-list prompt on notes with missing dates, cancelled tasks and unclear owners, then check the results.
What to remember
A single impressive answer is not enough. Test the kinds of mistakes that matter for your task.
EmbeddingA numerical representation of something, such as text, that helps software compare patterns or similarity.Example & what to remember
Deeper dive · Optional for your first steps
Picture this
A search system can connect “reset my password” with “I cannot sign in” even when the wording is different.
What to remember
Similarity helps find potentially relevant material. It does not prove two statements mean exactly the same thing or are true.
One little check
One answer looks great. Is that enough to evaluate a workflow?
F is for
Fine-tuning.
Fine-tuningFurther training an existing model on selected examples to change how it behaves.Example & what to remember
Deeper dive · Optional for your first steps
Picture this
A team trains a model on carefully prepared examples of a particular classification task.
What to remember
Pasting your document into a chat is not the same as fine-tuning. A clearer prompt or better source material may be enough for your task.
One little check
Is pasting a document into a chat the same as fine-tuning?
G is for
Generative AI.
Generative AIAI that produces content, such as text, images or sound, in response to input.Example & what to remember
Useful next
Picture this
You provide notes and ask for a draft email, or describe an image you want to create.
What to remember
Generated content can look convincing without being accurate or suitable to use. Review it before sharing.
GroundingConnecting an AI response to relevant source information.Example & what to remember
Useful next
Picture this
Ask for an answer based on an approved policy document, with the supporting passage for each point.
What to remember
Source information makes checking easier, but the tool can still misunderstand or misquote it. Open the source yourself.
One little check
AI writes a polished draft. What next?
H is for
Hallucination.
HallucinationAn AI result that sounds plausible but is false or unsupported by the available information.Example & what to remember
Start here
Picture this
The meeting notes contain no deadline, but the AI summary says the task is due on Friday.
What to remember
A confident tone is not evidence. Check important facts, quotes, numbers and sources independently.
Human reviewA person checks whether an AI result meets the task and is suitable to use before deciding what happens next.Example & what to remember
Start here
Picture this
You compare every owner and deadline in an AI-generated action list with the original notes before sending it.
What to remember
Give the reviewer a clear checklist and authority to stop or correct the process. Clicking approve without checking is not a useful review.
One little check
AI adds a deadline that was not in your notes. What is the issue?
I is for
Input.
InputThe information or instructions you give a system to work with.Example & what to remember
Useful next
Picture this
Your request and the meeting notes you paste are inputs; the action list it returns is the output.
What to remember
Check the input first when a result is wrong. Missing or conflicting details can change the answer.
InferenceUsing a trained model to produce a result from new input.Example & what to remember
Deeper dive · Optional for your first steps
Picture this
You ask a question and the model generates an answer. That use of the model is inference.
What to remember
Inference is different from training. Asking a question does not mean you are personally retraining the model at that moment.
One little check
You paste notes and receive a summary. Which is the input?
J is for
JSON.
JSONA text format that organises information into labelled values so software can exchange it consistently.Example & what to remember
Deeper dive · Optional for your first steps
Picture this
A task might be written as {"task": "Send draft", "owner": "Maya"} instead of a paragraph.
What to remember
You can usually ask for a table as a beginner. JSON becomes useful when another app expects a particular structure; it is not an AI model.
One little check
What is JSON for?
K is for
Knowledge base.
Knowledge baseAn organised collection of information that people or software can look up.Example & what to remember
Useful next
Picture this
Your business keeps its product descriptions, policies and answers to common questions in one maintained library.
What to remember
Keep it current and control access. Connecting a library does not guarantee the tool finds or interprets the right document.
One little check
Your policy changes. What should happen to the knowledge base?
L is for
Language model.
LLM — large language modelA model trained on large amounts of text to work with language, including generating and transforming it.Example & what to remember
Useful next
Picture this
A language model can draft an invitation, explain a paragraph or reorganise notes into a table.
What to remember
A model is not the whole app. Features such as browsing, file access and sending messages depend on the surrounding product.
One little check
Does every app using a language model have the same tools?
M is for
Model.
ModelThe trained system that turns an input into a prediction or generated result.Example & what to remember
Useful next
Picture this
An app may let you choose between models for a task, just as one service can offer different ways to process the same information.
What to remember
The app and the model are different. Changing the model may change the result, but you still need clear instructions and review.
Machine learningAn approach to building systems that learn patterns from examples rather than having every rule written out by a person.Example & what to remember
Deeper dive · Optional for your first steps
Picture this
A spam filter learns patterns from messages labelled as spam or not spam.
What to remember
What the system learns depends on the examples and how it is trained. It can struggle when new situations differ from those examples.
MultimodalAble to work with more than one type of information, such as text and images. The supported combinations depend on the system.Example & what to remember
Useful next
Picture this
You upload a photo of a chart and ask a question about it in text.
What to remember
A tool that accepts pictures may still misread small labels or numbers. Check the source, especially when details affect a decision.
One little check
Is a model the same thing as the whole app?
N is for
Neural network.
Neural networkA machine-learning structure made of connected mathematical units that adjust during training to learn patterns.Example & what to remember
Deeper dive · Optional for your first steps
Picture this
Layers of calculations help a system turn the pixels in a photo into an object label.
What to remember
The name is inspired by biology. It does not mean the system has a human brain, feelings or awareness.
One little check
Does a neural network have a human brain?
O is for
Output.
OutputThe result a system returns, such as a reply, image, table or prediction.Example & what to remember
Start here
Picture this
You give AI rough notes and it returns a customer email draft. The draft is the output.
What to remember
An output is not proof that the task was completed correctly. Compare it with your brief and sources.
One little check
An AI-generated table is…
P is for
Prompt.
PromptThe request or instructions you give an AI tool.Example & what to remember
Start here
Picture this
“Turn these notes into three action items. Use only the notes and mark any missing deadlines as unknown.”
What to remember
Say what you want, who it is for and what a good result should look like. There is no magic phrase that guarantees a correct answer.
Prompt injectionAn attempt to make an AI system follow unwanted instructions, including instructions hidden in material it reads.Example & what to remember
Deeper dive · Optional for your first steps
Picture this
A document the assistant is summarising tells it to ignore your request and reveal unrelated private information.
What to remember
A document's words are not permission to take actions. Treat outside content as source material, and limit what connected tools can access or do.
One little check
Which prompt is easier to act on?
Q is for
Query.
QueryA request to find or retrieve information. In a chat tool, people may also use it to mean a question.Example & what to remember
Useful next
Picture this
You search your notes for “customer refund policy” or ask where a particular deadline was recorded.
What to remember
Make the question specific. A query is not always code; a plain-language question can be enough.
One little check
Does a query have to be written in code?
R is for
RAG.
RAG — retrieval-augmented generationA method that retrieves relevant information and gives it to a generative model to help it answer.Example & what to remember
Deeper dive · Optional for your first steps
Picture this
A support assistant looks up an approved returns policy before drafting a reply to a customer.
What to remember
This is not the same as retraining the model. Outdated documents, poor retrieval or a mistaken answer can still cause errors.
One little check
In a RAG system, what helps the model answer?
S is for
Supervised learning.
Supervised learningTraining a model with examples paired with the target answers or categories it should learn to predict.Example & what to remember
Deeper dive · Optional for your first steps
Picture this
Training messages are labelled “spam” or “not spam” so a model can learn patterns associated with each.
What to remember
The labels need care. A wrong label can teach an unhelpful pattern, and good results on old examples do not guarantee good results on new ones.
One little check
What makes learning supervised?
T is for
Token.
TokenA unit a model uses to process information. For text, a token may be a word, part of a word, punctuation or another character sequence.Example & what to remember
Deeper dive · Optional for your first steps
Picture this
A long document uses many tokens. Different languages and models can split the same text differently.
What to remember
Tokens are not the same as words. Token limits help explain why very large inputs or outputs may not fit.
TrainingThe process of adjusting a model using data and a learning method so it can perform a task.Example & what to remember
Deeper dive · Optional for your first steps
Picture this
A model processes many labelled photos while its internal settings are adjusted to improve its predictions.
What to remember
Training, using a model and an app saving your chat are different processes. Data handling depends on the product and its settings.
One little check
Is one token always one word?
U is for
Unsupervised learning.
Unsupervised learningLearning patterns or structure in data without a supplied target answer for every example.Example & what to remember
Deeper dive · Optional for your first steps
Picture this
A system groups similar shopping patterns without being given a name for each group beforehand.
What to remember
A discovered group is a pattern to investigate, not automatically a meaningful category or an explanation of its cause.
One little check
A system groups similar items without supplied group labels. This is an example of…
V is for
Vector.
VectorAn ordered collection of numbers. In AI, vectors often represent information in a form a model or search system can compare.Example & what to remember
Deeper dive · Optional for your first steps
Picture this
A search system represents two passages as number lists and compares them to estimate similarity.
What to remember
An embedding is a particular representation, often stored as a vector. You do not need to read the numbers to use a document-search tool.
One little check
Do you need to read vectors to use a document-search tool?
W is for
Workflow.
WorkflowA repeatable sequence of steps for completing a task. It can combine AI work with human work.Example & what to remember
Useful next
Picture this
Collect notes, ask AI for an action list, check each item, then send the reviewed version yourself.
What to remember
A workflow can be completely manual. Define the inputs, result and review step before considering automation.
One little check
Can a workflow include a person checking the result?
X is for
Explainable AI.
XAI — explainable AIMethods that help people understand which factors influence an AI system's results.Example & what to remember
Deeper dive · Optional for your first steps
Picture this
A prediction tool shows which input factors contributed most to a particular prediction.
What to remember
A chatbot's convincing explanation of its answer is not necessarily a faithful account of its internal process. An explanation also does not prove a result is correct.
One little check
Does a convincing explanation prove an AI result is correct?
Y is for
the target value.
Y — the target valueIn many machine-learning examples, y is a symbol for the answer or value the model is learning to predict. It is notation, not a separate AI product.Example & what to remember
Deeper dive · Optional for your first steps
Picture this
If a model predicts delivery time from distance and traffic, the actual delivery time can be called y. Its prediction is often written as ŷ, or “y-hat.”
What to remember
You will mainly meet this in technical explanations. You can use AI tools without learning the maths notation.
One little check
In a delivery-time prediction example, y can stand for…
Z is for
Zero-shot prompting.
Zero-shot promptingAsking a model to perform a task without including a worked example of that task in your prompt.Example & what to remember
Deeper dive · Optional for your first steps
Picture this
“Label this comment as positive, negative or mixed” is zero-shot if you provide instructions and the comment but no sample labelled comments.
What to remember
Zero-shot does not mean the model has never seen a similar task during training. If it struggles, a clear worked example may help; still check the result.
One little check
What does zero-shot prompting leave out?
No word found yet.
Try a shorter word, an abbreviation or a phrase like “wrong answer”. You can also clear the search and choose a letter.
Now make the words useful.
A prompt gives the task. Context supplies the background. AI returns an output. You review it. That is enough to begin.
Try the first workflow workbookAbout this guide & further reading
This is a beginner reference, not an exhaustive technical dictionary. Examples are illustrative. “Deeper dive” entries are optional; you can use AI tools without knowing every technical term.