This post was originally published on this site
Imagine walking into a bar and buying a drink for the jukebox or tipping your toaster for perfectly browned bread. Sounds absurd, right? Yet, as AI tools continue to become more common in newsrooms and media companies, an interesting debate has emerged: Should we flatter AI chatbots with kindness or treat them like the cold, unfeeling algorithms they are?
The Case For Kindness
Being nice has its perks, even with AI chatbots. Research from Waseda University and the RIKEN Center suggests that polite prompts can significantly enhance the output of large language models. It might sound crazy, but just adding a simple “please” can improve the results of your chatbot conversation.
Key findings from this research include:
Performance Boost: GPT-3.5 achieved a 15.6% higher score with polite prompts in language understanding benchmarks. Reduced Bias: GPT-4 showed the lowest degree of bias with moderately polite prompts. Cultural Nuances: The impact of politeness varied across languages, reflecting cultural differences in communication norms. Creativity Enhancement: Polite prompts encouraged AI models to draw from a wider range of linguistic sources.
“These AI models are nothing more than sophisticated pattern recognition devices trained on swaths of polite and rude human interactions,” said Robert Caulk, CEO of Emergent Methods, via email. “As with any AI model, your input is pattern matched to what it saw during training; politeness matches thoughtful replies, rudeness matches apathetic replies.”
Courteous language often influences AI to mirror that behavior, resulting in more respectful exchanges. “When you are nice to an AI, it will probably be nice back