Watch: Gradient Descent

You just saw the math of learning: nudge a weight to lower the loss. Here it is in motion. Press play and watch a model roll down its loss curve toward the lowest point, the gradient (the slope) tells it which way is downhill, and the steps shrink as the ground flattens out near the bottom.

This is exactly what every w -= lr * grad update does, one step at a time. Watch it, step through it, then answer the check below.