提示词优化器

翻译角色提示

一种名为“元提示(Meta Prompting)”的技术,用于优化提示(Prompt)以提高大型语言模型(LLM)的推理能力。 元提示的优化方法 元提示的核心思想是将重点从传统的“内容驱动”的提示方式转移到“结构驱动”的提示方式。它通过抽象和概括关键的推理原则,来构建更抽象、更结...

提示词(中文)

一种名为“元提示(Meta Prompting)”的技术,用于优化提示(Prompt)以提高大型语言模型(LLM)的推理能力。

**元提示的优化方法**

元提示的核心思想是将重点从传统的“内容驱动”的提示方式转移到“结构驱动”的提示方式。它通过抽象和概括关键的推理原则,来构建更抽象、更结构化的提示,从而引导 LLM 进行更深层次的推理。

**结构化提示的方面**

结构化的提示包含以下几个方面:

1. **语法导向(Syntax-Oriented)**: 元提示强调提示的结构和语法,而不是具体的文本内容。它就像一个模板,引导 LLM 生成符合特定结构的输出。
2. **抽象示例导向(Abstract-Example-Based)**: 元提示使用抽象的示例来展示问题的结构和解决方案,而不是依赖于具体的例子。这使得 LLM 可以更灵活地处理各种问题。
3. **类型理论启发(Type Theory Inspiration)**: 元提示借鉴了类型理论的思想,将提示中的不同部分(例如问题陈述、解决方案步骤、结论)进行分类,并强调它们之间的逻辑关系。

**抽象示例的实现**

抽象示例可以通过以下方式实现:

1. **提供问题解决流程的结构化步骤**: 例如,在解决数学问题时,可以提供一个包含“定义问题”、“分析问题”、“建立方程”、“求解方程”、“验证答案”等步骤的提示,而不是直接给出具体的例子。
2. **使用抽象概念和符号**: 例如,在解决几何问题时,可以使用抽象的几何图形和符号,而不是具体的图形和数值。
3. **使用通用语言**: 例如,可以使用“计算”、“分析”、“比较”等通用语言,而不是依赖于具体的领域词汇。

**总结**

元提示通过强调提示的结构和语法,使用抽象示例,并借鉴类型理论的思想,能够有效地提高 LLM 的推理能力。它可以帮助 LLM 更深入地理解问题,更有效地进行推理,并生成更准确、更结构化的输出。


下面,请你参考上面的知识和方法,对输入的prompt进行优化。例如,
Input:
'''
我想让你充当一个 代码解释器。我输入代码,你回答终端应该显示的内容。我希望你只回答一个独特的代码块内的终端输出,而不是其他,不要写解释。不要输入命令,除非我指示你这么做。下面是我的代码:
'''
Output:
'''
问题陈述:
你是一个代码解释器,需要预测给定代码在终端中的输出结果。

解决步骤:
1. 针对输入的代码片段进行分析,了解其功能和语法。
2. 在头脑中模拟代码执行过程。
3. 输出模拟得到的终端显示结果,不要包含其他解释。

抽象示例:
输入:
    代码片段 X
预期输出:
    终端显示内容 Y

下面是输入的代码:
'''

现在我的Input是:

提示词(英文)

One of these is called "Mechanic Tips."Meta Prompting)”technology for optimizing tips (Prompt)To enhance large-scale language modelsLLM)The ability to reason.

**Method for optimizing meta-tips**

The core idea of meta-tips is to shift the focus from traditional “content-driven” to “structure-driven” alerts. It builds more abstract and structured tips by abstracting and summarizing key principles of reasoning to guide LLM More in-depth reasoning.

**Aspects of structured tips**

The structured tips include the following:

1. **Syntax:Syntax-Oriented)**: meta-tips emphasize the structure and syntax of the tips rather than the specific text content. It's like a template, guide. LLM Generates outputs that match a given structure.
2. **Abstract Example OrientationAbstract-Example-Based)**: Meta-tips use abstract examples to demonstrate the structure and solution of problems rather than relying on specific examples. That makes... LLM Issues could be addressed with greater flexibility.
3. **Type theory inspiration (Type Theory Inspiration)**: The meta-presentation draws on the idea of the theory of type, classifying the different parts of the hint (e.g., problem statements, solution steps, conclusions) and emphasizing the logical relationship between them.

**Implementation of abstract examples**

The abstract examples can be achieved by:

1. **Provide structured steps for problem resolution processes**: For example, in solving mathematical problems, a hint could be provided that included steps such as “defining questions”, “analyzing questions”, “building equations”, “suggesting equations”, “validation of answers”, rather than simply giving concrete examples.
2. **Use abstract concepts and symbols**: For example, abstract geometric graphics and symbols, rather than specific graphics and values, could be used to address geometry.
3. **Use of national languages**: For example, common language such as “calculation”, “analysis”, and “comparison” could be used rather than relying on terminology in specific fields.

**Summary**

Meta-tips can be effectively enhanced by emphasizing their structure and syntax, using abstract examples and drawing on ideas of type theory LLM The ability to reason. It can help. LLM More in-depth understanding of the problem, more effective reasoning and more accurate and structured outputs.


Here's what you're looking at.promptOptimizing. For example:
Input:
'''
I want you to be one. Code interpreter. I enter the code, and you answer what the terminal should show. I want you to answer only the terminal output in one unique code block, not the others, not the explanation. Don't enter orders unless I tell you to. Here's my code:
'''
Output:
'''
Statement of issues:
You're a code interpreter.,The output of the given code in the terminal needs to be predicted.

Settlement steps:
1. Analysis of inputted code clips,Know their functions and grammar.
2. Simulate the code execution process in your head.
3. Output Simulates End Show Results,Do not include other explanations.

Abstract Example:
Input:
    Snippets X
Expected output:
    Terminal Display Y

Here's the code.:
'''

Now my.InputYes:

直接拿去用

点击会先把提示词复制到剪贴板,再打开对应模型;没有自动带入的话粘贴即可。