In-context learning is proof that LLMs are not stochastic parrots.