To expand of that: the whole idea is that normalization is not applied after every step. This approach is efficient when you know you have to add a bunch of nu

To expand of that: the whole idea is that normalization is not applied after every step. This approach is efficient when you know you have to add a bunch of numbers together. You add them up, and then normalize. The example can be, e.g. doing long multiplication, or adding all numbers in a list. (Just repeating what was said in the parent comment for emphasis)