Get comfortable · Lesson 1 of 16

What Is ChatGPT? A Simple Explanation

What ChatGPT is, how it writes an answer one word at a time, what it does well, and why you should still double-check what it tells you.

Plays from YouTube. Nothing loads from YouTube until you press play (privacy).Watch on YouTube

English captions available in the player. Prefer reading? Follow the written lesson below.

THE WRITTEN LESSONFollow along at your pace

Everyone is talking about ChatGPT, but it is not always clear what it is. This lesson explains three things in plain words: what ChatGPT is, how it writes its answers, and when you can trust it. There is nothing to click yet. Most pictures are simple drawings from the video, and the two in the birthday example show a real chat in ChatGPT.

What ChatGPT is

ChatGPT is a chat assistant. It is a computer program you talk to by typing, just like sending a message to a friend. You write in plain, everyday English. There are no special commands to learn. You just ask, and it writes a reply.

On the left, a phone shows the message Can you help me plan a picnic? and the reply Sure! Pack sandwiches, fruit and a blanket, then pick a sunny spot by the lake. On the right, the heading A chat assistant and the line A computer program you talk to by typing, above a ticked label that says Plain, everyday English and a crossed-out label that says Special commands. — If you can send a text message, you already know the basic idea.

You may also hear this kind of program called a chatbot. Because it creates new text when you ask, ChatGPT is also a kind of generative AI.

An example: a birthday message

Say you want help with a birthday message. You type:

Type this

Can you help me write a birthday message for my mom?

A few seconds later, a warm, friendly message appears. Here, ChatGPT puts the message in its own box, with a copy button at the top right. You can copy it, change it, or use it as it is.

ChatGPT in a browser window at chatgpt.com. The question Can you help me write a birthday message for my mom? is partly cut off at the top. The reply opens with Of course! and puts the message in a box labelled Writing, starting Happy Birthday, Mom! and ending I love you so much, Mom. Happy Birthday! The copy button at the top right of the box is outlined, with a label that says Copy it with one click. Under the box, ChatGPT offers to make the message more emotional, short and sweet, or more personal, and asks one question about you and your mom. — One click on the copy button copies the whole message.

And it is a conversation. If the answer is not quite right, just say so: make it shorter, make it funnier, or add her name. In the video, the follow-up is:

Then reply

Make it shorter and add her name: Maria.

ChatGPT remembers what you said earlier in the same chat, and it changes the answer. The new message is shorter and opens with “Happy Birthday, Mom Maria!”

The same chat, further down. The reply Make it shorter and add her name: Maria. sits in a light blue bubble on the right, under a label that says It remembers this chat. Below it, a shorter message in a Writing box starts with the outlined line Happy Birthday, Mom Maria! and ends Love you so much, Mom! Under the box are two suggestions: Make it sound more natural and Shorten it to two sentences. — The new answer builds on the first one, because both are in the same chat.

Under the new message, ChatGPT suggests more changes you could ask for, such as “Make it sound more natural” or “Shorten it to two sentences”.

How it writes: one word at a time

So how does ChatGPT write its answers? One word at a time. It looks at everything written so far and picks a likely next word. Then it does it again, and again, very fast.

Try it yourself. Finish this sentence: The cat sat on the… If “mat” came to mind, you just picked a likely next word. Sofa, floor or roof could fit too, but they are less likely.

The words The, cat, sat, on and the, each in a white box, followed by mat in a green box. Below, a bar chart titled Likely next words, where the bar for mat is much taller than the bars for sofa, floor and roof. On the right, a round arrow above a ticked label that says Again, and again, very fast. — Mat has the tallest bar, so mat fills the gap.

How does it know which word is likely? It learned from a huge amount of text: books, articles and websites. By reading all of that, it learned how people usually write. That’s why it can write about almost anything.

Drawings of two books, a newspaper article and a web page, labelled Books, Articles and Websites, with dashed arrows that lead into one large speech bubble. Under the bubble, a ticked label says How people usually write, and a green label at the bottom says It can write about almost anything. — All that reading is where its sense of a likely word comes from.

What it is great at: word work

This makes ChatGPT great at word work:

Drafting emails and messages.
Explaining hard things in simple words.
Brainstorming ideas, like gift ideas or names for a project.
Summarizing long text into a few short points.

Four cards under the title What it's great at: Drafting emails, with an envelope; Explaining simply, with a light bulb; Brainstorming, with a gift box and two small light bulbs; and Summarizing, with a long page next to a short list of three points. — All four jobs are made of words, and words are what it learned from.

The catch: it can sound sure and be wrong

But here’s the catch. ChatGPT is built to write words that sound right. That is not the same as being right. It can sound very sure, and still be wrong. It can even make up a fact, a date or a quote.

A card that says Sounds right, with a green tick, and a card that says Is right, with a question mark, on either side of a not-equal sign. Below, a speech bubble says Yes, I am completely sure., with a red cross at its corner. At the bottom, three warning labels: a fact, a date, a quote. — A sure tone is not proof that an answer is right.

For example, it might tell you the Eiffel Tower was finished in 1899. It sounds confident. But the real year is 1889.

A box that says The Eiffel Tower was finished in, with 1899 crossed out in red and 1889 in green with a tick, labelled The real year. Under the box, a warning label says Sounds confident. — Only one digit is wrong, yet the answer sounds completely sure.

Two simple habits

So build two simple habits:

Double-check. For health, money, legal questions and important dates, always check what it tells you with a trusted source.
Keep private details out. Don’t type private details, like passwords or bank numbers, into the chat.

Two cards. The first, Double-check, shows small drawings for Health, Money, Legal and Dates, then a magnifying glass and the words A trusted source. The second, Keep private details out, shows a padlock labelled Private with a red cross, and a password box and a bank card, each struck through in red, labelled Passwords and Bank numbers. — Check the important things, and keep the private things to yourself.

For a closer look at what AI assistants do well and where they go wrong, read AI assistants for beginners.

In short

ChatGPT drafts, explains, brainstorms and summarizes. It can sound sure and still be wrong, so check health, money, legal questions and dates with a trusted source, and keep private details out of the chat.

A recap on a green background, headed Remember, with three ticked points: A fast writing partner (drafts, explains, brainstorms, summarizes), Not an all-knowing expert (it can sound sure and still be wrong) and Check what matters (health, money, legal, dates, private details). A white card on the right says Try it: Pick one writing job for ChatGPT to help with. — Three things to remember, and one small job to try before the next lesson.

Before the next lesson, think of one small writing job you could ask ChatGPT to help with.

YOUR TURN

In the next lesson, you will sign up and take a tour of the screen: How to sign up for ChatGPT and find your way around.

Read the full video transcript

Everyone is talking about ChatGPT. Maybe you've heard the name, but you're not sure what it really is. That's fine. In this lesson, you'll learn three things. What ChatGPT is. How it writes its answers. And when you can trust it.

ChatGPT is a chat assistant. It's a computer program you talk to by typing, just like sending a message to a friend. You write in plain, everyday English. There are no special commands to learn. You just ask, and it writes a reply.

Here's an example. You type: Can you help me write a birthday message for my mom? A few seconds later, a warm, friendly message appears. You can copy it, change it, or use it as it is.

And it's a conversation. If the answer isn't quite right, just say so. Make it shorter. Make it funnier. Add her name. ChatGPT remembers what you said earlier in the same chat, and it changes the answer.

So how does it write? One word at a time. It looks at everything written so far, and picks a likely next word. Then it does it again, and again, very fast. Try it yourself. The cat sat on the... mat.

How does it know which word is likely? It learned from a huge amount of text. Books. Articles. Websites. By reading all of that, it learned how people usually write. That's why it can write about almost anything.

This makes it great at word work. Drafting emails and messages. Explaining hard things in simple words. Brainstorming ideas, like gift ideas or names for a project. And summarizing long text into a few short points.

But here's the catch. ChatGPT is built to write words that sound right. That isn't the same as being right. It can sound very sure, and still be wrong. It can even make up a fact, a date, or a quote.

For example, it might tell you the Eiffel Tower was finished in eighteen ninety-nine. It sounds confident. But the real year is eighteen eighty-nine.

So build two simple habits. First, for health, money, legal questions, and important dates, always double-check with a trusted source. Second, don't type private details, like passwords or bank numbers, into the chat.

So think of ChatGPT as a fast, helpful writing partner. Not an all-knowing expert. You bring the question and the judgment. It brings the words. Before the next lesson, think of one small writing job you could ask it to help with. Next, you'll sign up and take a tour of the screen.