第三大脑

https://effectivesoftwaredesign.com/2023/05/21/prompt-engineering-is-not-a-profession/

Prompt Engineering is not a profession. Prompt Engineering is a skill. It is the ability to write effective prompts for Generative Models like ChatGPT, and it can be potentially useful in many professions.
提示工程不是一种职业。及时工程是一种技能。它是为 ChatGPT 等生成模型编写有效提示的能力，它在许多职业中都有潜在的用处。

In the same way that millions of professionals use Word and Excel in their daily jobs, people will need to write prompts for Generative Models to accomplish their tasks. But this does not mean that there will be a category of specialists that will be paid to spend most of their time writing good prompts. If you don’t like the example of Microsoft Office, we could also say that ChatGPT is only a tool, in the same way that Figma, Jira and Git are tools.
就像数以百万计的专业人员在日常工作中使用Word和Excel一样，人们需要为生成模型编写提示来完成他们的任务。但这并不意味着会有一类专家会花大部分时间写好的提示。如果你不喜欢Microsoft Office的例子，我们也可以说ChatGPT只是一个工具，就像Figma，Jira和Git是工具一样。

But in my opinion it is even more important to understand that the effective usage of Large Language Models (LLMs) such as ChatGPT requires Domain Knowledge. A professional must have previous experience with the domain in order to write good questions, and also to be able to validate that the responses produced by the model really make sense.
但在我看来，更重要的是要了解像ChatGPT这样的大型语言模型（LLM）的有效使用需要领域知识。专业人士必须具有该领域的经验，才能写出好的问题，并且能够验证模型生成的响应是否真正有意义。

A person must have previous Domain Knowledge to be able to write a really effective prompt in this domain. So, for example, if you would like to ask ChatGPT to produce an amazing travel itinerary to some touristic destination, you must include in your prompt the important aspects you are looking for. If you are not able to determine the building blocks that compose a great travel itinerary, you will not be able to optimize your prompt.
一个人必须具有以前的领域知识才能在这个领域编写真正有效的提示。因此，例如，如果您想要求 ChatGPT 制作前往某个旅游目的地的惊人旅行行程，您必须在提示中包含您正在寻找的重要方面。如果您无法确定构成出色旅行行程的构建块，您将无法优化提示。

Besides that, Domain Knowledge is required to understand if the response produced by the model is correct and complete. It is a well-known fact that LLMs can hallucinate, generating answers that sound plausible but actually are plain wrong. Very often also the results produced by these models are incomplete, missing important information. Only a person that has experience with the domain will be able to identify if the text that was generated by the model is wrong or incomplete.
除此之外，还需要领域知识来了解模型生成的响应是否正确和完整。众所周知，LLM可以产生幻觉，产生听起来合理但实际上是完全错误的答案。通常，这些模型产生的结果也不完整，缺少重要信息。只有具有域经验的人员才能确定模型生成的文本是错误的还是不完整的。

Therefore, the conclusion is that there will not be such a thing as a generalist Prompt Engineer. In many professions, people will be required to learn how to write good prompts. The professionals who will become the best prompt writers will combine their ability to use Generative Models with their previous Domain Knowledge.
因此，结论是不会有通才提示工程师这样的东西。在许多职业中，人们将被要求学习如何编写好的提示。将成为最佳提示作者的专业人士将把他们使用生成模型的能力与他们以前的领域知识结合起来。