What Is Generative AI?

Introduction

You have probably seen chatbots that write emails, explain homework or help with code. They are examples of generative AI: software that creates new text, images, audio or code from a short request called a prompt.

The simple explanation

A language model is a very large pattern machine that predicts what text should come next.

Imagine your phone suggesting the next word as you type. A language model does a much bigger version of that. It was trained on huge amounts of text, so it learned which words and ideas tend to follow each other. When you ask a question, it builds an answer one small piece (called a token) at a time, choosing likely next pieces.

This is why its answers sound fluent. It is also why it can sound confident when it is wrong: it is producing likely text, not looking up a verified fact.

Some words you will meet

• Prompt: what you type to ask for something.
• Token: a small chunk of text, often a word or part of a word.
• Training: the stage where the model learns patterns from lots of data.
• Context window: how much text the model can keep in view at once, including your messages and its replies. Anything outside it is forgotten.

What AI is good at

• Explaining an idea in simpler words, or in several different ways
• Drafting and rewriting text, such as an email or a summary
• Brainstorming options and examples
• Helping you practise, for example by quizzing you

Where you need to be careful

• Facts, numbers, quotes, dates and links can be wrong or invented
• It may not know recent events unless it can search
• It can repeat bias found in its training data
• It does not know your private situation unless you tell it, and you should be careful what you share (see the last lesson)

Try it yourself

Think of one topic you are learning. Ask a chatbot to explain it "like I am ten". Then ask for one example. Notice what feels clear, and one thing you would want to double check somewhere else.

Summary

• Generative AI creates new content from a prompt.
• A language model predicts likely next text, so answers are fluent but not guaranteed true.
• It is great for explaining, drafting and brainstorming, and risky for exact facts without checking.