Differences in the Moral Foundations of Large Language Models