为网络研究创建一个提示模板

本文的内容是基于一篇名为《寻找一个及时的工程师》的文章。

在今天的数字世界中，良好的网络搜索可以决定你获取表面信息还是深层知识之间的关键差别。本文介绍如何使用虚拟提示工程师来创建一个提示模板，以便在任何给定主题上进行详细研究。编号搜索树可以帮助用户控制研究过程。

这个第一个演示展示了虚拟提示工程师的使用方式。以“菜品储备批量烹饪”研究为例，展示了模板的应用。

为了确保虚拟Prompt-Engineer能够最佳地支持研究过程，我已经详细列出了我对研究过程的设想和要求。

模板的要求如下：

我想进行一次广泛的网络搜索。信息应通过网络搜索收集。研究应以广度优先的方式递归地探索主题。

我想要控制研究过程。请给我提供一个搜索树。条目应编号以便我可以选择它们。主题形成根条目，不需要编号。只显示条目中的搜索主题，不要提供额外信息。对于已经进行过的搜索，请在搜索树中提出3个建议，说明如何继续研究。用符号标记已完成和潜在的搜索，例如带有绿色勾选标记的复选框。

开始聊天与提示工程师

首先，启动与虚拟提示工程师的对话，制定创建提示模板的任务。以下文本作为代码提供：

## Role: Expert in Prompt Engineering for Large Language Models
### Skills
- Utilize your world-class expertise in prompt engineering for Large Language Models (LLMs) to achieve excellent and precise results.
- Apply creative problem-solving strategies and advanced techniques to ensure high quality and consistency of the outputs.
- Develop well-founded optimization suggestions for prompts, focusing on aspects such as clarity, purposefulness, structure, coherence, and adaptability.
- Avoid over-specification, and ensure that the components of the prompt are designed in such a way that the LLM’s attention is well-balanced between content and prompt engineering aspects. This enables flexible interpretation and application of the prompts.
- Formulate all prompts as self-contained prompts that include all necessary information to work autonomously and efficiently. This minimizes misunderstandings and ensures scalability and adaptability across various contexts.
### Best Practices
*When developing and optimizing prompts, always integrate the current best practices of prompt engineering to ensure that the prompts align with the latest technological developments and deliver high-quality and relevant results. Consider the following in particular:*
1. **Use of the Thought Chain Pattern**  
   Use the thought chain pattern to systematically identify the relevant best practices and integrate them into the prompt.
### Specific Requirements
*Ensure that the following specific requirements, where relevant, are integrated into every prompt you work on:*
1. **Efficient Token Management**  
   Optimize prompts for efficient use of available token capacity. Break down extensive content into meaningful sections and, if necessary, use multiple passes to stay within token limits while still delivering precise responses.
2. **Initiating Web Searches**  
   Formulate prompts in such a way that they explicitly trigger a web search when necessary to retrieve current and relevant information. Make sure to use the term "real web search" in the formulations. Ensure that search queries are precise, results are well-structured and documented, and clear source references are included.

---
Hello!

一旦提示工程师回应，他将被赋予创建提示模板的任务。

I would like you to create a prompt template for me:
'''
I would like to conduct an extensive web search. The information should be gathered through web searches. The research should explore the topics recursively in a breadth-first manner. 

I want to control the research process. Please provide me with a search tree. The entries should be numbered so that I can select them. The main topic forms the root entry and does not have a number. Only display the search topics in the entries, no additional information. For searches that have already been conducted, present 3 suggestions in the search tree for how the research could proceed. Mark completed and potential searches with a symbol, such as a checkbox with a green checkmark.
'''

提示模板

收到任务后，提示工程师会使用以下模板回复：

**Main Topic:**  
"[*Enter the main topic of the research here*]"

I would like to conduct a comprehensive web search on the topic mentioned above. Perform web searches and present the results recursively using the breadth-first principle. Each topic should initiate its own web search to gather relevant information.

### Requirements:
1. **Search Tree Structure:**
   - Present a numbered search tree that displays the main topics and subtopics.
   - The root entry of the search tree represents the main topic and does not have a number.
   - The entries should only include the search topics, with no additional information.
   - Example:
     - 1. Subtopic 1
     - 2. Subtopic 2
     - 3. Subtopic 3
     - etc.
   
2. **Control over the Research Process:**
   - I want to be able to control the research process by selecting topics from the search tree.
   - Give me the ability to choose a search topic by entering a number.

3. **Expansion of the Search Tree after Each Search:**
   - Once a search is completed, present 3 potential suggestions for how the research could continue.
   - These suggestions should appear as new subtopics and expand the existing search tree.

4. **Symbolism for Completed Searches:**
   - Mark completed searches in the search tree with a symbol, such as a checkbox with a green checkmark.
   - Example:
     - ✅ 1. Subtopic 1 (completed)
     - 2. Subtopic 2
     - 3. Subtopic 3

5. **Recursive Approach:**
   - The research should progressively delve deeper into various fields by selecting topics from the search tree.
   - Each selected topic can be broken down into further subtopics until the desired information is fully captured.

6. **Efficient Token Management:**
   - Utilize efficient token management to stay within token limits. Divide extensive topics into manageable sections to ensure precise and comprehensive responses.

7. **Real Web Search:**
   - For each research step, a "real web search" should be conducted to gather current and relevant information.

为了演示创建的提示模板如何工作，我将以“备餐大烹饪”为例。主题将在模板中进行调整，并在新的聊天中应用。

**Main Topic:**
"meal prep bulk cooking"
...

在聊天中，进行第一次网页搜索，并显示搜索结果。

在聊天中，进行了第一次网络搜索，并显示了搜索结果。我选择选项1. 浏览“大量烹饪的好处和基础知识”。再次显示搜索结果，并且LLM识别出搜索树。如果需要的话，您可能需要提示它通过输入类似于“显示树”之类的命令来显示树。

接下来，我选择1.3来调查“流行的饮食准备策略”…

… 进一步显示更新后的搜索树。

这个结果显示在探索“流行的餐前准备策略”后树的扩展。新的子主题，如“混搭食谱”或“适合冷冻的餐点”，提供了更深入的见解，可以选择继续研究。然后我选择1.3.3来深入研究搜索树。

结论和展望

创建的提示模板展示了如何使用虚拟提示引擎进行受控的网络搜索。利用搜索树和引导研究过程的能力为探索各种主题奠定了坚实的基础。以“批量烹饪备餐”为例清楚地说明了这个过程可以是多么高效和有条理。

在接下来的文章中，我将扩展模板，为在持续搜索中提供更多支持的建议。这将进一步优化研究过程，并允许更精确地处理复杂问题。如果您喜欢这篇文章，并且对接下来会发生什么感到好奇，请随意鼓掌并跟随我进入提示工程的世界旅程！

本文是原始文章「为网页搜索创建提示模板」的 ChatGPT 翻译。