What is AI? A plain-English starter

Learn in your browser · Chapter 01 of 06

AI comes with a pile of confusing vocabulary. Five ideas will get you most of the way.

Artificial intelligence (AI) is the broad name for computer systems that do things which normally need human intelligence: understanding language, recognising pictures, making recommendations, planning. It is a big umbrella, not one technology.

Machine learning is one way of building AI. Instead of a programmer writing every rule by hand, the computer is shown a very large number of examples and works out the patterns for itself. Spam filters and photo tagging work this way.

Deep learning is machine learning that uses layered networks of simple mathematical units (neural networks), loosely inspired by the brain. Almost every headline AI system of the last decade is built on deep learning.

Generative AI is AI that creates new content: text, images, audio, video or computer code. When people say "AI" today they usually mean generative AI.

A large language model (LLM) is the kind of generative AI that works with text. It powers most chatbots and writing assistants. The rest of this workbook is mostly about LLMs.

Scroll sideways to see every column.

Learn in your browser · Chapter 02 of 06

You do not need any maths to understand the core idea. It is simpler, and stranger, than most people expect.

A large language model is trained on an enormous amount of text. During training it plays one game, billions of times: given some text, predict what comes next. Each time it guesses wrong, its internal numbers are nudged so that next time it guesses a little better.

Text is chopped into small pieces called tokens. A token is often a short word or part of a word. As a rough rule of thumb in English, a token is about three quarters of a word. The model does not read letters like you do. It works with these pieces.

When you send a message, the model looks at your prompt (everything you have typed, plus any hidden instructions) and predicts a likely next token. It adds that token, then predicts the next one, and the next, until it has written a complete reply. That is all a chatbot is doing: repeatedly choosing a likely next piece of text.

After the first training stage, models are trained further to follow instructions and to be helpful and polite. That is why they answer questions rather than just continuing your sentence.

Two more terms will help. The context window is how much text the model can take into account at one time: your messages, its replies and any documents you add. Anything outside the window is invisible to it. And a model has a knowledge cutoff: it learned from text up to a certain date, so on its own it may not know about recent events. Some tools add web search to help with this.

Answers can vary when a language model samples from possible next tokens. Some decoding settings instead choose the most likely next token. The actual reply depends on the model, its settings and the supplied context. Repeated identical replies do not prove an answer is correct.

Learn in your browser · Chapter 03 of 06

Knowing the strengths is how you get real value quickly.

Language models are strongest at work that is about language and pattern, where a good first draft is valuable and you can judge the result yourself.

• Drafting and rewriting: emails, letters, reports, social posts, in a tone you choose.
• Summarising: turning a long document or meeting notes into a short list.
• Explaining: describing a difficult idea at the level you ask for.
• Brainstorming: giving you twenty options to react to, so you are never staring at a blank page.
• Translating and changing format: turning notes into a table, or a paragraph into bullet points.
• Helping with code: explaining errors, writing small pieces, suggesting fixes.

Notice the pattern. In each case you stay in charge. AI gives you a fast starting point and you decide whether it is good enough.

Learn in your browser · Chapter 04 of 06

Regular users tend to meet the same handful of problems. Learn them now and they will not catch you out.

A hallucination is when the model states something that is false or invented as if it were true. It can invent a quotation, a statistic, a book title or a legal case that does not exist. It is not lying. It is producing plausible text.

• Exact facts and figures: names, dates, statistics and quotes can be wrong.
• Sources: a model may produce a reference that looks real but is not.
• Arithmetic: long sums can be wrong unless the tool uses a calculator.
• Recent events: without search, it may not know what happened lately.
• Its own certainty: it is often just as confident when it is wrong.

1. Ask yourself: if this is wrong, does it matter? The higher the stakes, the more checking you need.
2. Check any name, date, number or quote against a trustworthy source you find yourself.
3. If it gives a source, open it and confirm it exists and says what the model claims.
4. Ask the model to list what it is unsure about, or to answer again in a different way. Compare.
5. For health, legal, money or safety decisions, use a qualified person. AI can help you prepare questions, not replace them.

Learn in your browser · Chapter 05 of 06

A few habits, built now, protect you and other people later.

Anything you type into an online AI tool is sent to a company's computers. Different tools handle that data differently, and the rules can change, so read the privacy settings of the tool you use.

• Never paste passwords, bank details, private keys or other secrets.
• Do not paste other people's personal information without a good reason and permission.
• Be careful with anything confidential for your work or your customers. Check your organisation's rules first.
• Look in the settings for options about whether your conversations are used to improve the service, and choose what you are comfortable with.
• Remember that you are responsible for what you publish or send, even if AI helped write it.

AI can also run on your own computer or inside your organisation. A local application may still use cloud models or connected services, so its location alone does not prove that information stays inside. Check the actual configuration, receiving services and your organisation's handling rules. Later workbooks explore these arrangements.

Learn in your browser · Chapter 06 of 06

Use a free chatbot and non-sensitive text. Each experiment takes about ten minutes.

Well done. You now understand the basics of how AI works. The next workbook, Your first prompts, teaches you how to ask for exactly what you want.