AI School › How it’s built inside · Lesson 66 of 130

🎯 Training, explained with numbers

AI learns by guessing, failing and correcting itself many times.

Imagine learning to sink a basketball shot blindfolded. Your first throw misses by a meter. A friend says "too far left!" and you shift your aim a bit. You throw again, miss by half a meter, adjust again. After a thousand throws, you barely miss. AI learns exactly like this, but with numbers. It tries to guess an answer, checks how far off it was from the right one, then slightly changes its internal numbers to miss a little less next time. That "how far off I was" is called the error.

The difference is that AI does not make a thousand tries, but billions, super fast, one after another. Each time its error shrinks a little, like that basket getting more accurate. Nobody feeds it the answers by heart: it is shown tons of examples and left to adjust on its own, step by step. It is a bit like walking down a hill in the fog, always stepping downward: you cannot see the bottom, but if you keep going down you eventually reach it. At the end of this long training, the AI's numbers are so well tuned that it guesses right almost every time.

🎒 Try it in class (or at home)

Try tossing a paper ball into a bin: adjust your aim after each throw. Count how many tries until you stop missing.

← Lesson 65: Why so much power is needed · Lesson 67: "Weights" made easy →

📱 Download the Android app (beta) iOS coming soon

Say what you mean. Get what you need.

Grace Certified, the AI coach that trains and certifies your prompt engineering, by Agora Intelligence.