Rude is the new smart

Based on this article by Alan Bradley at Live Science.

A new study reveals a surprising twist in our interaction with chatbots: being rude to advanced AI models like ChatGPT-4o can actually increase their accuracy. Researchers found that the most "very rude" prompts yielded a higher success rate (84.8%) on test questions than the most polite ones (80.8%), defying the assumption that politeness improves performance.

While this finding is scientifically intriguing, the authors strongly warn against adopting a toxic tone, emphasizing that the benefit doesn't outweigh the negative effects on user experience and communication norms. Ultimately, the study highlights how sensitive LLMs remain to even superficial cues, creating a bizarre trade-off between performance and well-being.

Check out this article.

Related Stories