Does differential privacy make PAC learning much harder?

Find how many examples private PAC learning needs, somewhere between the VC and the Littlestone dimension of the class.