The Butterfly Effect of Altering Prompts: How Small Changes and Jailbreaks Affect Large Language Model Performance