The task: split the two teams
Below are two teams of dots. The AI's whole job is to draw ONE line that puts blue on one side and pink on the other. Right now its line is a bad guess. Each wrong dot is a mistake — and "mistakes" added up is called the loss.
Team Blue
Team Pink
✗ = a dot on the wrong side
Loss
–
Accuracy
– %
Step
0 / 30
How big are the mistakes? (shorter = better)
weight a
–
weight b
–
bias
–
Press Train ▶ and watch the line wiggle into place — 30 tiny nudges, loss dropping each step. It runs exactly the same every time (no luck involved).
This is exactly what "training" means.
The model didn't get a brain transplant — it just nudged its 3 numbers (weight a, weight b, bias) a tiny bit at a time, each step lowering the loss. That's it. Real AI models like ChatGPT do the very same dance — only with billions of numbers instead of 3. Same idea, bigger orchestra. 🎻