The Dangers of Stochastic Parrots: Can Language Models Be Too Big?