🏔️ Build a mini-brain: how AI learns
AI 'learns' by tuning weights so the error gets smaller. Move the weight yourself and find the valley (minimum error). Then hit 'let it learn' — and watch the model do the exact same thing automatically.
Weight: -2.40Error: 5.210 steps
Goal: roll the ball to the lowest point — that's where the error is smallest.
🐢 slowrisky ⚡
💡 This is how neural networks learn
This is 'gradient descent': the model checks which way the error drops and takes a small step that way. Too high a learning rate — it overshoots the valley and bounces (or blows up). Too low — it learns forever. Real AI does exactly this, just with millions of weights at once.