Researchers at the University of Southern California have found that conversational robots like ChatGPT perform much better when addressed politely. — AFP Relaxnews
“Compose a message to customer service about a lost package”, “Can you rewrite this paragraph to make it more concise?”... It’s not always easy to know how to talk to ChatGPT. However researchers at the University of Southern California have found that conversational robots like ChatGPT perform much better when addressed politely.
Abel Salinas and Fred Morstatter came to this conclusion after testing different “prompts” (written requests) with several large language models, including ChatGPT. The academics, for example, added spaces at the beginning and end of their requests, or punctuated them with polite formulas.
It turns out that the way you address a chatbot has an impact on the accuracy of the query. It's best to start your “prompt” with a “hello” and end it with a “thank you”. In fact, a user who ignores the elementary rules of politeness will not obtain the same result as one who respects them.
In contrast, Salinas and Morstatter have found that it’s pointless trying to bribe an artificial intelligence. ChatGPT reacts badly to bribe attempts: it doesn’t produce better quality answers when you try to motivate it with a small tip.
And it doesn’t get any better when the amount offered is larger – quite the contrary. If a user promises US$1,000 (RM4,778) in exchange for a “perfect” answer, OpenAI's conversational robot performs even less effectively.
Its competitor, LlaMA, however, seems a little more sensitive to the prospect of a monetary reward. “When it comes to influencing the model by specifying a tip versus specifying we will not tip, we found that tipping US$1 (RM4.70), US$10 (RM47) or US$100 (RM477) to LlaMA-7B significantly improves the performance,” Salinas and Morstatter note in their study.
“On some online forums it makes sense for someone to add a greeting, like Quora, for example. Starting with ‘hello’ or adding a ‘thank you’ is common there,” explains Salinas in a statement, adding that these conversational elements could be shaping the process of learning language patterns.
As a general rule, it’s advisable to think carefully about the way you communicate with a conversational artificial intelligence. But there’s no need to overcomplicate your “prompts”: simplicity is good, especially with ChatGPT. – AFP Relaxnews