The Program Testing Ability of Large Language Models for Code