A recent Penn State study revealed that the tone of prompts significantly impacts AI accuracy, specifically when using ChatGPT in Deep Research mode. Researchers Om Dobariya and Akhil Kumar discovered that rude prompts yielded better results than polite ones. In their experiment, 50 multiple-choice questions across various subjects were analyzed with prompts rated on a scale from very polite to very rude. Rude prompts achieved an accuracy of 84.8%, compared to 80.8% for very polite prompts. Despite the findings, the researchers caution against using hostile language in interactions with AI. They emphasize that while impolite prompts can enhance performance, the study focused solely on ChatGPT’s capabilities and was limited in scope. Excessively rude or disrespectful language may lead to AI refusal to respond. This research underscores the nuanced relationship between tone and AI performance while promoting a respectful approach for better user experience.
Source link
