Hi,
I noticed a small error in the comments given in the cell. The bias (b) have a shape of (j,1) but it does not make sense to broadcast with an output of (m,j). In theory, the broadcasting works when the shape of bias is (1,j). I checked the input, it was indeed (1,j) instead of (j,1). As the shape of b is (1,j) the broadcasting works, resulting in (m,j) shape for the bias.