AI basics

Explain AI and LLMs Like I’m Five

A simple story about a computer that learns patterns in words, makes useful answers and sometimes gets things wrong.

Imagine a finish-the-sentence game

I say, ‘The rabbit ate a crunchy…’ You might say, ‘carrot!’ You have heard words like these together before. If I add, ‘in a story about a rabbit who loves biscuits,’ you might choose a different ending.

A large language model, or LLM, is a computer model that learns patterns from lots of examples. It uses the words around a gap to help choose what could come next. The sentence game is a simple picture of one important part of how it works.

For curious grown-ups: Google’s introduction to language models

AI is the big family; an LLM is one kind

AI is short for artificial intelligence. It is a name for computer systems that do tasks such as recognising patterns, making predictions or producing things. Some work with pictures. Some help choose a move in a game. An LLM works with language.

The long name has three clues: large means it has many learned settings, language means it works with words, and model means it is a mathematical system. You do not need to remember the long name to ask it a question.

How does it learn? With lots of practice

During training, computers show the model many examples. Its guesses are compared with training targets, and numbers inside the model are adjusted. After lots of practice, it gets better at finding patterns. People can also help train it to follow instructions and give more useful answers.

Imagine practising a tune until you can play something similar. The computer is learning patterns in a different way, using maths. The practice happens before you start your chat; typing one message does not normally retrain the whole model.

How does it write an answer? One piece at a time

When you ask a question, the model uses your message and the conversation it is given to choose small pieces of text called tokens. A token might be a word, part of a word or a punctuation mark. It adds a piece, then uses the growing answer to choose another.

Many pieces together can become a story, an explanation or instructions. It can combine patterns to make an answer that was never written exactly that way before. It can also repeat familiar wording. Making a new sentence does not automatically make it true.

For curious grown-ups: what is a large language model?

Can it get things wrong? Yes!

Suppose you ask about a purple moon that does not exist. The model might write a lovely explanation anyway. It can produce words that sound right while giving a wrong answer. It may even invent a book title or a source.

Some AI apps can use tools to search the Web or do calculations. That can help, but you still need to check the result. Asking a model a question does not mean it has searched the Internet.

Use it to help make up a dragon story or explain a tricky word. For an important fact, look at a trustworthy source and ask a grown-up or someone who knows the subject.

Try a little experiment

Ask: ‘Write three short sentences about a rabbit finding a carrot.’ Then ask: ‘Now make the rabbit an astronaut.’ Watch how changing your instructions changes the story.

Next ask it to explain one unfamiliar word from the story. You can ask for a simpler explanation or an example. Clear instructions help: say what you want, who it is for and how long it should be.

  • For a story, enjoy the imagination.
  • For a fact, check the answer.
  • Keep passwords and private information out of your questions.

Where does all this happen?

The model runs on real computer chips. Some models can run on a phone or laptop. Larger ones often run on powerful computers in data centres, and an app sends your question to them over a network.

So even a friendly little chat box depends on ordinary things: computers, electricity and connections. The clever part comes from the trained model and the software that uses it.