Comment by yarky

The problem with preconceptions about your parameters is that you might be missing some crazy cool path to your goal, which you might find by randomly exploring your sample space. I remember seeing this same principle in mcmc samplers using uniform priors. Why is this so crazy?