The problem with Perceptrons was that it appeared to prematurely dismiss multi-layer networks, but the underlying insight about the limits of connectionist architectures remains valid even as networks have become much deeper and more sophisticated.