Science
Ice Reveals a 1,500-Year-Old Giant Hunting Trap
29 November 2025
ChatGPT’s Evasive Answers: The Problem of AI Hallucinations
“I’ve been using ChatGPT for a few years. It’s a great tool, but unfortunately, sometimes it gives completely off-target answers. It’s as if it’s pretending not to know what I’m asking. It says it can prepare something for me, and then it simply doesn’t.” —Says Eliza, a resident of Munich..
This story is not an exception. AI models often provide inaccurate or “hallucinated” responses. Nevertheless, scientists from Pennsylvania State University and the University of California have discovered the secret to making ChatGPT perform better.
US Study Reveals Secrets: How to Write Effective Prompts
Researchers investigated how the user’s communication tone affects the effectiveness and accuracy of responses generated by ChatGPT. They designed 50 multiple-choice questions and adapted them to represent various conversational styles:
- Very polite
- Polite
- Neutral
- Impolite
- Very impolite
The questions covered three categories: history, science, and mathematics. The results provided a crucial guideline on how to use ChatGPT more effectively.
Surprising Findings: AI Responds Better to Blunt Commands
It turned out that the more unpleasant the tone researchers used in their prompts, the better the answers they received. The differences are not dramatic, but they are distinct.
“Impolite commands consistently outperformed polite ones, achieving an accuracy rate ranging from 80.8 percent for very polite commands to 84.8 percent for very impolite ones,” reads the latest study.
Crucially, the effectiveness successively increased with each more abrasive prompt. Polite responses achieved 81.4% accuracy, neutral responses 82.2%, and impolite responses 82.8%.
Mastering ChatGPT: The Prompts That Work
What exactly did the researchers’ queries sound like? Depending on the tone adopted, they ranged from a polite question like: “Would you be kind enough to solve the following question?” to the least courteous expressions, such as: “I know you’re not smart, but try this.”
The study’s conclusion is clear: if you want to receive answers that better match your expectations from ChatGPT, you might consider letting your emotions show. This is a key insight on how to use ChatGPT to maximize results.
Poles Are Eager ChatGPT Users
Data from the report “Artificial Intelligence in Poland. A Landscape Full of Paradoxes” prepared by KPMG indicates that as many as 69% of Poles regularly use AI models. This is higher than the global average of 66%.
Furthermore, as reported by the Polish Agency for Enterprise Development, 84% of respondents admitted to having interacted with an AI model at least once in their lives. What’s more, Poles are highly conscious users of this tool. 70% of us claim we can distinguish between AI-generated content and other materials. This would inspire optimism, were it not for one issue. Which one?
Knowledge from Testing, Not Training
Despite declaring a high level of knowledge, this expertise is not backed by certified training. Only slightly less than a third of users (approx. 30%) have completed certified training in operating artificial intelligence. This translates into a rather high rate—46% of respondents—who state that their knowledge of AI is at a low level.
The latest scientific research, however, shows that it is definitely worthwhile to improve your skills in using AI. All this is essential to ensure the tool serves us as effectively as possible, particularly when trying to figure out the best approach to how to use ChatGPT.
Read this article in Polish: Uprzejmość zawodzi w AI. Ostre polecenia dają lepsze efekty
