Multi-Prompt Collaboration: Re-understanding Prompts and Their Application Misunderstandings
Introduction
With the deepening of practical applications of large models, many users of large models and Prompt creators have become increasingly proficient in using large models. Various learning materials and high-quality content related to Prompts continue to emerge. As practices related to Prompts advance, people’s understanding of Prompts has become deeper. However, it has also been found that many friends still have many misunderstandings in the process of creating prompts and using the capabilities of large models. There is still a lack of high-quality content related to multi-Prompts collaboration. This article is thus written with the hope of helping everyone use large models more skillfully.
When introducing multi-Prompt collaboration, let’s re-understand what a Prompt is, starting from the most obvious yet most overlooked aspect: What is a Prompt?
What is a Prompt?
It is well-known that a Prompt is a tool for invoking the capabilities of large models. However, in the actual use of large models, there are many overlooked aspects in the understanding of Prompts.
Misunderstandings in Understanding Prompts
Let’s start with a common misunderstanding about Prompts.
Many people think that a Prompt is a prompt word obtained from various platforms and used at the beginning of a ChatGPT conversation, similar to this:
The following prompt is from the well-known ChatGPT Chinese Training Guide project ([6]):
I hope you can act as an English translator, spelling proofreader, and rhetoric improver. I will communicate with you in any language, and you will identify the language, translate it, and reply to me in more beautiful and concise English. Please replace my simple words and sentences with more elegant and refined expressions, ensuring that the meaning remains unchanged but making it more literary. Please only answer the corrected and improved part without writing explanations. My first sentence is “how are you ?”, please translate it.
This kind of prompt word used at the beginning of a conversation often requires prompt engineers to invest a lot of time and energy in creating and iterating. Since such prompt words take effect at the system level, and for API application development, they are used in the System part, we will refer to this kind of system-level prompt word as System Prompt (the prompt word at the beginning of the conversation or the content of the System prompt word) hereafter.
Because many tutorials, articles, and Prompt sharing platforms share such Prompts, many friends’ understanding of Prompts is only limited to System Prompts, and they even simply equate Prompts with System Prompts.System Prompt is very important, especially for prompt engineers. However, from the user’s perspective, especially when dealing with tasks that require multi-turn continuous conversations to solve, simply using a good System Prompt is far from enough for making good use of large models.