Prediction

Imagine someone says to you: "The sky is ____." What word comes next?

You'd probably say "blue." Maybe "clear" or "dark" or "falling", but "blue" feels like the most natural answer.

That's exactly what AI does. It reads a bunch of words and guesses what comes next. That's the whole trick. Seriously, the most powerful AI systems in the world are, at their core, just predicting the next word over and over again.

How did it get so good at guessing? It learned by reading billions of sentences from books, websites, articles, and more, that's its "database" of language patterns. After all that reading, it picked up on which words tend to follow which.

No magic. No thinking. Just a really, really good guessing machine.

When the AI sees "The sky is," it doesn't just pick one word. It looks at ALL the words it knows, thousands and thousands of them, and gives each one a score.

"Blue" gets a high score. "Green" gets a lower score. "Refrigerator" gets a very low score (because skies usually aren't refrigerators).

Think of it like a bar chart. Each word has a bar, and taller bars mean the AI thinks that word is more likely to come next. The AI then picks the word with the tallest bar.

Look at the bar chart below, you can see which words the AI thinks are most likely for "The sky is ____."

Now it's your turn. Below you'll see a sentence with a blank. Tap on the word you think the AI would pick. The bars show how confident the AI is about each option, taller bars mean more likely.

After you pick, you'll see if you matched the AI's top choice. Then hit "Try another sentence" to see more examples.

Notice something: you're actually really good at this! Your brain predicts words all the time, when you read, when you listen, when you finish someone's sentence. AI is doing the same thing, just with math instead of neurons.

Sometimes you want the AI to be creative. Sometimes you want it to be precise.

Imagine you're turning a dial. Turn it down, and the AI becomes very focused, it almost always picks the most likely word. Turn it up, and the AI becomes more adventurous, sometimes picking surprising words.

This dial is called temperature.

Low temperature: "The sky is blue." (safe, predictable)
High temperature: "The sky is weeping." (creative, unexpected)

Try the slider below, drag it from "Focused" to "Creative" and watch how the bars change. Low temperature makes one word dominate. High temperature spreads the bars out, giving the AI more options.

You now know the core idea behind every AI language model: predict the next word.

One important thing to understand: the AI doesn't store all those sentences it read during training like a database you can search. Instead, it compressed what it learned into millions of numbers called weights. The training data is gone, what's left is the patterns.

But here's a question, when we say the AI looks at "words," what does it actually see? It can't read letters the way you do. It doesn't see "blue" and think "that's a color."

It turns out, AI doesn't see words at all. It sees numbers.

In the next lesson, we'll learn how text gets turned into something a computer can actually work with. Get ready to meet tokens.