Thinking about Network Size

$\frac{0\mathrm{.}0}{1\mathrm{.}0}$ point $($graded$)$

What is the danger to having too many hidden units in your network?

It will take up more memory

It may overfit the training data

It will take longer to train