Home / AI Courses / Inside the AI Brain: How LLMs Actually Think and Learn! / [Lesson 3/5] How the Robot Learns: From Wild Reader to Polite Assistant!
🤖 Young AI Explorers Series · Part 3

How the Robot Learns: From Wild Reader to Polite Assistant!

A brand new AI starts with zero knowledge—like a newborn baby robot! How does it learn all human science, and why did it have to go to Robot Etiquette School? Let's find out!

📚 1. Step 1: The Great Library Marathon (Pre-training)

Imagine a super-fast speed-reader with millions of eyeballs reading every book, encyclopedia, recipe, and cartoon on the internet. As the robot reads, it plays an endless guessing game: "Guess what word comes next!"

Whenever it guesses wrong, it calculates its mistake score (the Loss) and slides down the "Error Mountain" by turning its billions of digital knobs!

🛷 Toy 1: Sliding Down the Error Mountain (Gradient Descent)

Click the button to train the robot and watch its mistake score shrink:

🎓 2. Step 2: Robot Etiquette School (Fine-Tuning & Manners)

After reading the whole library, the robot knows everything. But there is a funny problem: It has no manners! If you say: "Can you write a poem about autumn?", it might just keep babbling: "Autumn is followed by winter. Winter has snow. Snow is white...". It doesn't realize you asked it to STOP and write a poem for you!

🦁 The Wild Genius (Base Model)

Read the entire web, but only knows how to continue typing endlessly. Doesn't know when to be helpful or polite.

🌟 The Polite Assistant (Aligned Model)

Went to Etiquette School! Human teachers gave it gold stars for helpful, polite, and safe responses. Now it acts as your friendly co-pilot!

← Previous: 2. Attention Magic 📚 Course Overview Next Lesson: 4. Word Dominoes →