Incorporating negative emotional stimuli into prompts can significantly improve the performance of large language models across a variety of tasks.