Part 2/6:
Studies have demonstrated that treating AI models with kindness and framing instructions politely can significantly enhance their accuracy and efficiency. For example, a recent experiment involving Google's PaLM model—a language model similar to ChatGPT—showed remarkable results when users adopted a collaborative tone.
When users instructed the AI to "take a deep breath and go step by step," the model's accuracy soared to about 80%. In stark contrast, ignoring such framing and simply issuing commands resulted in a lower accuracy of 34%. This suggests that positive reinforcement and respectful communication can effectively guide AI systems to perform tasks more reliably.