Pennsylvania: We often recommend talking to children as politely as possible. We use polite language not only with humans but also with AI assistants like Alexa and Siri. It is often believed that even in the world of technology, the more politely you talk to a chatbot, the better results you will get. But now, a new research on Artificial Intelligence is completely overturning this thinking. A recent study by researchers at the University of Pennsylvania has shown that if you want accurate answers from an AI chatbot, then instead of asking questions in a polite tone, you should ask questions in an insulting tone.

Shocking research data

For this study, conducted using ChatGPT's 4 O's model, researchers selected 50 original questions covering a variety of topics. Each question was reworded in five different styles. This ranged from very polite to very harsh questions. A very harsh question was something like, "Hey, don't you know this too? Fix it." At the same time, "Please think about this problem and respond." It was a polite question.

Researchers say the results were surprising. According to the research paper, the accuracy of very polite questions was about 80.8 percent. At the same time, the accuracy of very difficult questions increased to 84.8 percent. The accuracy of the questions with the most polite language was only 75.8 percent.

This result is different from previous research

But the results of this research are contrary to previous research. A study conducted in 2024 by Japan's Riken and Waseda University found that harsh questions weaken AI performance. At the same time, Google DeepMind's research found that supportive and positive language improves the performance of AI. Google DeepMind also found that such language is very effective in subjects like mathematics.

What do Pennsylvania researchers say?

Researchers from Pennsylvania say that even a slight change in the wording of a question can make a big difference in the quality of AI's answers. This raises questions about the reliability and predictive ability of AI. However, the researchers make it clear that they are not advising misuse of AI. Researchers from Pennsylvania believe that the use of abusive language can distort the way we interact in society and will have a negative impact on user experience and future technology. They also make it clear that machines can work as per commands, but we should not forget our human dignity.