Extrapolating outside the latent space distribution for GANs?

Hello everyone, I was wondering what happens when you extrapolate out of the latent space distribution (noise vector) for a Generative adversarial network (GAN). Can anybody help me?