Introduction

In the traditional model of generative AI interaction, the process is quite straightforward: you ask a question, and the language model (LLM) responds. This approach works well when the user knows exactly what they need to ask. However, what happens when the user doesn't know where to start or asks a question without providing the necessary background information? This is where the flipped interaction technique comes into play.

Imagine you're seeking advice from an investment advisor. In a typical scenario, you might ask, "What should I invest in?" While this question is a good starting point, it lacks the context needed for the advisor to provide tailored advice. The advisor would need to know more about your financial goals, risk tolerance, investment horizon, and other relevant details.

Now, let's apply this to an LLM. Instead of waiting for the user to ask the perfect question, the LLM can take the initiative to ask questions that help gather the necessary background information. For example, the LLM might ask, "What are your investment goals?" or "How much risk are you willing to take?" By doing so, the LLM can better understand the user's needs and provide more accurate and personalized advice.

This white paper explores how the flipped interaction prompt engineering technique can enhance the effectiveness of large language models (LLMs), using the example of an investment advisor. It demonstrates how an LLM can ask targeted questions to better understand a user's investment goals and background before providing tailored advice. This technique improves the quality of interactions and ensures users receive the most relevant and helpful information.