Just to elaborate on what TheEd said (because understanding the concept means it's easier to remember):
The idea of a diagonalised matrix is that it takes a two particular vectors (the eigenvectors) and scales them.
Now, as it is, the matrix
(2332) tells you that it sends
(10) to
(23) and
(01) to
(32). However, you also know that it sends
(11) to
5×(11) and
(−11) to
−1×(−11), so what you want to get is a matrix which:
1. Sends
(10) to
(11) and
(01) to
(−11) (i.e. it sends the basis vectors to the eigenvectors)
2. Scales these two vectors appropriately
3. Takes you back to where you started
Since these both do the same thing to both the basis vectors, they must be the same thing, so they're equal.
This is why when you diagonalise you get something in the form
A=P−1DP, where
D is diagonal: the matrix
P sends your basis vectors to the eigenvectors of the matrix, then
D scales them, and finally
P−1 takes you back to where you were.