Moral Foundations of Large Language Models