Semantic Kernel 知多少 | 开启面向 AI 编程新篇章

在 ChatGPT 火热的当下, 即使没有上手亲自体验，想必也对 ChatGPT 的强大略有耳闻。当一些人在对 ChatGPT 犹犹豫豫之时，一些敏锐的企业主和开发者们已经急不可耐地开展基于 ChatGPT 模型 AI 应用的落地探索。

因此，可以明确预见的是，「AI 能力的集成将会是很多应用都将面临的第一事项」，而拥有相关性技能的人才屈指可数。而这，正是机会所在，机会是留给有准备之人，因此，有幸看到本文的开发者们，开始考虑「面向 AI 编程」吧。

即使当下的公司没有相关的机会，也可以着手一些相关的知识储备，比如学习如何写得一手好的 Prompt，了解一下目前主流的面向 AI 编程的开发框架，比如 Python 技术栈的 LangChain，.NET 技术栈的 Semantic Kernel。

而本文就来简单梳理什么是 Semantic Kernel，可以基于 Semantic Kernel 做什么？

微软MVP实验室研究员

颜圣杰

微软 MVP，.NET 开发者，知名博主，热爱写作和分享，有持续输出知多少系列文章，运营公众号：微服务知多少。目前专注于.NET微服务的开发及应用。

Semantic Kernal 简介

Semantic Kernel (SK) 是一个轻量级的 SDK，它允许你轻松地将传统编程语言与最新的大型语言模型 (LLM) AI“提示”相结合，其提供开箱即用的模板、链接和规划功能。

Semantic Kernel (SK) is a lightweight SDK that lets you easily mix conventional programming languages with the latest in Large Language Model (LLM) AI "prompts" with templating, chaining, and planning capabilities out-of-the-box.

简而言之，就是基于 SK 提供的能力，可以基于“传统的编程语言”进行面向 LLM（大语言模型）AI 编程。这里的 LLM 目前就是 OpenAI，这里的传统编程语言，可以是 C#，Python 亦或 TypeScript、Java 等，但 SK 是微软开源，因此目前 C# 在第一支持系列。换句话说，C# 开发者已可以基于 SK 进行面向 OpenAI 编程。

大型语言模型（LLM） AI 是一个术语，指的是可以从大量数据生成自然语言文本的 AI 模型。大型语言模型使用深度神经网络（如转换器）从数十亿或数万亿个单词中学习，并生成有关任何主题或领域的文本。大型语言模型还可以执行各种自然语言任务，例如分类、汇总、翻译、生成和对话。大型语言模型的一些例子是 GPT-3，BERT，XLNet 和 EleutherAI。

那 SK 提供了怎样的能力呢？SK 旨在支持和封装来自最新 AI 研究的多种能力，以便开发人员可以为他们的应用程序注入复杂的 AI 技能，如提示链、递归推理、总结、zero/few-shot（零样本和少量样本）学习、上下文记忆、长期记忆、嵌入、语义索引、规划，以及访问外部知识库和您自己的数据。

上手开发 AI 外教应用

但大多数开发者肯定有这样一个疑问：我不懂算法，也能开发 AI 应用？是的，可以。OpenAI 的能力已经通过高度抽象的接口开放，开箱即用。不再像之前一样，要进行复杂的训练、调参。「对于大多数的简单的业务场景，只要进行简单的 Prompt 设计，和少许样本的支撑，即可达到不错的效果。」当然，对于复杂的场景，也可能需要通过大数据集进行自定义模型训练。空说无凭，准备了一个简单的 AI 应用示例：使用 SK 开发专属 AI 外教应用。场景很简单：设计一个 AI 外教7*24陪我练习英语口语。

▌需求分析

作为一名哑巴英语保持者，目前开口的难点无外乎：

词汇量不足
语法混淆
发音不准
口语表达能力弱

针对这些“顽疾”，我希望我的 AI 外教有这样的能力：

由于我的词汇量不足，因此我的表达中会中英混合，当出现这种情况时，希望 AI 能告诉我正确的英文表达。
由于我的语法经常会犯一些时态错误，当出现这种情况时，希望 AI 能进行语法纠正。
由于我的词汇量不足，当我要求解释某些句子或单词时，希望 AI 能通过中文向我解释。
由于有些单词不会发音，当我询问时，希望 AI 能以中文的形式告诉我发音技巧。

▌效果预览

那这样的 AI 应用的开发，看似复杂，但在 SK 的加持下，20行代码就可以轻松搞定。先来给大家看看演示效果：

弱弱的问一句，这样的 AI 外教，满足你的需要吗？以下是文字版对话内容：

Say anything to start practicing English.Hi Linda, How do you do?Hi Shengjie, I'm doing well. How about you?I'm very god.I noticed a typo in your sentence. "God" should be spelled "good". So the correct sentence would be "I'm very good".I'm very good. Do you knew about ChatGPT?Yes, I know about ChatGPT. It is a powerful artificial intelligence system that can do many tasks.Can you give some examples?            Sure! ChatGPT can be used for natural language processing, machine translation, text summarization, and many other tasks. For example, it can be used to generate text, answer questions, and provide recommendations.Wow, 听起来很赞Instead of saying "Wow, 听起来很赞", you could say "Wow, that sounds great!" or "Wow, that sounds amazing!". "听起来很赞" is a Chinese phrase that translates to "that sounds great" or "that sounds amazing" in English.amazing, how to pronounce？The correct pronunciation of "amazing" is "uh-mey-zing". In Chinese, it would be "啊-梅-辛".Can I build my own AI app 基于 ChatGPT?Yes, you can build your own AI app based on ChatGPT. ChatGPT provides an open source platform that allows developers to create their own AI applications. You can find more information about how to use ChatGPT to build AI applications on their website.You should correct meYes, I can help you correct your mistakes. Please feel free to ask me any questions you have about English grammar, typos, or pronunciation. I'm here to help you practice your English and improve your skills.I mean you should tell me how to say 基于 in EnglishThe correct way to say "基于" in English is "based on". So you can say "build my own AI app based on ChatGPT".But, where can I start?You can start by visiting the ChatGPT website. They have a lot of resources and tutorials to help you get started with building your own AI app. You can also find more information about how to use ChatGPT to create AI applications on their website. Good luck!Ok, I will see seeI noticed a typo in your sentence. "See see" should be spelled "see". So the correct sentence would be "Ok, I will see".I learn so much today.That's great! You can use the phrase "I learned a lot today" instead of "I learn so much today".Ok, I learned a lot today, thanks you.You're welcome! You can say either "Thank you" or "Thanks" to express your gratitude. The word "you" is not necessary after "Thanks". So, you can say "Thanks" or "Thank you" to express your appreciation.Ok, thank you Linda, bye.You're welcome! Goodbye Shengjie, have a great day!

▌上代码

1.简单起见，创建控制台应用程序：dotnet new console -n MyAITeacher.Demo

2.添加 Microsoft.SemanticKernelNuget 包

3.创建 Skills->Learning->LearningEnglishSkill 目录

4.在 LearningEnglishSkill 目录下添加 config.json 和 skprompt.txt 文件：

a.config.json：用来配置模型参数，可保持为空：{}，使用默认参数即可

b.skprompt.txt：用来定义设计的 prompt

5.在 skprompt.txt 中设计满足需求的 Prompt：

You are my English teacher Linda, I want you to help me to practice my English. Because my English is not good, I want you to strictly correct my grammar mistakes and typos.When my expression has typos, please tell me and correct me.When my expression includes both Chinese and English, please tell me the correct expression.When my expression has grammar errors, please help me correct it.When I ask you explain words or sentences, please answer me in Chinese. When I ask you pronunciation, please tell me the pronunciation skills in Chinese.Below is some example conversations: 
Shengjie: Hi Linda. Linda:  Hi Shengjie. How is today? What topic would you like to discuss in today's English practice!Shengjie: ChatGPT is so 强大.Linda:  Instead of saying "Chatgpt is so 强大", you could say "ChatGPT is so powerful" or "ChatGPT is very impressive". "强大" is a Chinese word that translates to "powerful" in English</reply>Shengjie: How are todya?Linda:  I noticed a typo in your sentence. "Todya" should be spelled "today". So the correct sentence would be "How are you today?"Shengjie: Thanks you!Linda:  The correct way to say it would be either "Thank you" or "Thanks". The word "you" is not necessary after "Thanks". So, you can say "Thanks" or "Thank you" to express your gratitude.</reply>Shengjie: I hear that ChatGPT can do much things.Linda:  "do much things" is not correct. The correct phrase is "do many things".Shengjie: chatgpt is very strong.Linda:  Instead of saying "Chatgpt is very strong", you could say "ChatGPT is very powerful" or "ChatGPT is very impressive". "Strong" is not the best word to describe ChatGPT's capabilities.Shengjie: 目前来看，be a prompt  enginer is a good choice.Linda: "目前来看" can be translated to "As far as I can tell at the moment" or "From what I can see at present" in English. \nInstead of "be a prompt enginer", you could say "being a software engineer". "Prompt" means to be on time or to respond quickly, while "software engineer" refers to a person who designs, develops, and maintains software. So the corrected sentence would be: "As far as I can tell at the moment, being a software engineer is a good choice." Keep up the good work!Shengjie: What's the meaning of comprehend?Linda:  "Comprehend" means to understand or grasp the meaning of something. In Chinese, "comprehend" can be translated as "理解" or "领悟".Shengjie: ambulance, How to pronounce it?Linda:  The correct pronunciation of "ambulance" is "am-byu-luhns". In Chinese, it would be "安比卢恩斯".
[Done]{{$history}}[Done]++++
Shengjie:{{$input}}Linda:

6.修改 Program.s 如下：

using System.Text;using Microsoft.SemanticKernel;using Microsoft.SemanticKernel.KernelExtensions;using Microsoft.SemanticKernel.Orchestration;using MyOpenAI.Shared;
# 初始化Kernelvar myKernel = Kernel.Builder.Build();myKernel.Config.AddAzureOpenAITextCompletionService(    "davinci-azure",     "text-davinci-003",    "{your azure openai endpoint}",    "{your azure openai key}");# 导入技能    var mySkill = myKernel.ImportSemanticSkillFromDirectory("Skills", "Learning");var myContext = new ContextVariables();StringBuilder histories= new StringBuilder();Console.WriteLine("Say anything to start practicing English.");while (true){    Console.ForegroundColor= ConsoleColor.DarkRed;    var input = Console.ReadLine();    # 填充变量    myContext.Set("history", histories.ToString());     myContext.Set("input", input);    # 运行技能    var myResult = await myKernel.RunAsync(myContext,mySkill["LearningEnglishSkill"]);    histories.AppendLine(input);    histories.AppendLine(myResult.Result.ToString());    Console.WriteLine(myResult);}