We trained them on billions of words from the internet (books, websites, chats). They predict the next word like super-powered autocomplete. Example: You type “Why is the sky…” → it guesses “blue?” because it’s seen that question a million times.

It’s how most modern AI “thinks.” Imagine layers of tiny math switches that turn on or off based on what they see. Billions of them working together copy how brain cells connect. Simple version: It’s like a huge team of yes/no guessers passing notes until they agree on an answer.

Machine learning is the main way AI gets smart. Instead of programming every rule, we let the computer find patterns in data by itself. Analogy: Like teaching a kid to ride a bike by letting them fall a bunch of times instead of explaining physics.

Almost all AI today is “narrow” — great at one thing only (like translating languages or beating you at chess). There’s no “super intelligent” AI that can do everything a human can. Example: Google Translate = narrow AI. It can’t also drive a car.

Computers learn by looking at tons of examples (we call this “training data”). The more good examples, the better they get. Example: To recognize dogs, show the computer 10,000 pictures of dogs. It learns what “dog” looks like.