Investigating the Effects of Cognitive Biases in Prompts on Large Language Model Outputs