Whose Side Are You On? Investigating the Political Stance of Large Language Models