Course 4, lesson 31 of 100, Ages 10+
What’s a model?
The recipe AI writes for itself
Like I’m 5
A model is like a recipe that AI writes for itself after lots of practice. Then it follows that recipe to answer new questions.
The big idea
A model is what an AI ends up with after training: a big set of numbers that turns inputs into outputs. Show it a photo, and out comes 'cat'. Give it a sentence, and out comes a translation.
Training is like a cook adjusting a recipe after every tasting. The examples are the tastings, and the numbers in the model are the amounts of each ingredient. Once trained, the model can follow its recipe on brand-new inputs.
Examples
- Spam model: Input an email, output 'spam' or 'not spam'.
- Price model: Input house size and location, output an estimated price.
- Language model: Input some words, output the next likely word.
Check your understanding
- What is an AI model?
- Options: What the AI learned, saved as numbers; A toy robot; A photo of a computer.
Answer: What the AI learned, saved as numbers. A model is the learned recipe, stored as lots of numbers. - What is a trained model mostly made of?
- Options: Lots of numbers learned from examples; Cogs and springs; A list of every possible answer.
Answer: Lots of numbers learned from examples. Training sets the model's numbers so it can handle new inputs.
Remember
A model is what AI keeps after learning: lots and lots of numbers.
Talk about it
If you wrote a recipe for being a good friend, what would be in it?
Go deeper
Those numbers are called parameters. Training adjusts them to reduce mistakes. Small models have thousands; the biggest have billions.