In exercise number 5, I don't understand the usefulness of changing the loss function, if we then continue to calculate the weights and biases in the same way.

to update the values in each epoch

Follow this topic