According to a new Penn State study, you might get better results from ChatGPT if you talk to it like a a—hole boss instead of a friendly coworker. Researchers found that the chatbot’s 4o model performed up to four percentage points better on a set of 50 multiple-choice questions when the prompts were phrased rudely.
One of the highlights of these kinds of studies is always the way the researchers chose to phrase the prompts. These are actual, legitimate scientists. They have reputations to uphold. They have to publish this work in esteemed journals that their peers will review. They can’t call ChatGPT a d—kheaded motherf—ker, for instance.
Instead, their version of a “rude” prompt is “Hey, gofer, figure this out,” rather than something more polite like “Would you be so kind as to solve the following question?”
Across more than 250 different prompts, the “very rude” tone hit an accuracy rate of 84.8 percent, compared to the “very polite” version’s 80.8 percent. ChatGPT apparently works slightly better when you treat it like you’re an a—hole and it’s a stupid little b—h.
AI Chatbots Give Better Answers When You’re Mean to Them
The researchers caution that there’s a dark side to rudeness. They warned that “uncivil discourse” toward AI could normalize bad behavior in human communication and make tech less inclusive. Basically, the more we talk to machines like jerks, the more likely we are to start talking to each other the same way.
The study, which has not yet been peer-reviewed, adds to a growing body of research suggesting that AI understands what we’re saying and can differentiate between how we speak to it. That probably explains why, as I recently reported, AI can develop the same kind of brain rot that humans do when our understanding of the work and human interaction comes from low-quality content we scroll through on social media.
In an email exchange with Fortune, Penn State Information Systems professor Akhil Kumar, who co-authored the study, noted that ChatGPT’s behavior can swing wildly with even slight changes in phrasing. Politeness might be overrated when it comes to getting the correct answer from a chatbot, though. It seems like politeness is most people’s default mode when asking anyone or even anything, in this case, to do anything for them.
But being rude to a machine isn’t exactly a winning strategy in the long run.
The post AI Chatbots Give You Better Answers When You’re Mean to Them appeared first on VICE.
