Want Better Results From an AI Chatbot? Be a Jerk

A recent study from Penn State reveals that rude prompts can yield better results from AI chatbots compared to polite ones. Researchers discovered that very rude prompts produced correct responses 84.8% of the time, while very polite ones achieved 80.8%. This finding challenges previous studies which suggested that politeness enhanced AI performance. The researchers examined 250 prompts, ranging from very polite to very rude, across various subjects like math and science. The study implies that tone may affect AI response accuracy significantly, warranting a reconsideration of prompt engineering. Unlike past findings that viewed tone as a neutral factor, this study suggests that straightforward language might clarify intent, thereby improving the AI’s performance. Though the paper has yet to undergo peer review, its findings have already sparked interest among researchers in AI interaction techniques. The implications extend beyond etiquette, as they raise questions about the objectivity of AI outputs and how different tones influence results.

Source 🔗