Name: 人工智能代理，一目瞭然 (AI Agents, Clearly Explained)
Uploaded: 2025-04-24T03:10:05.000Z
Duration: 10 min 9 s

程度副詞

All right, most explanations of AI agents is either too technical or too basic.

好吧，大多數關於人工智能代理的解釋要麼過於專業，要麼過於基礎。

基礎被動語態

This video is meant for people like myself.

這段視頻就是為我這樣的人準備的。

You have zero technical background, but you use AI tools regularly, and you want to learn just enough about AI agents to see how it affects you.

你沒有任何技術背景，但你經常使用人工智能工具，而且你想了解人工智能代理，看看它對你有什麼影響。

In this video, we'll follow a simple one, two, three learning path by building on concepts you already understand, like chatGPT, and then moving on to AI workflows, and then finally AI agents, all the while using examples you will actually encounter in real life.

在本視頻中，我們將遵循簡單的一、二、三學習路徑，以您已經理解的概念為基礎，如 chatGPT，然後轉向人工智能工作流，最後是人工智能代理，同時使用您在現實生活中實際遇到的例子。

And believe me when I tell you those intimidating terms you see everywhere, like RAG, RAG, or React, they're a lot simpler than you think.

相信我，當我告訴你那些你隨處可見的令人生畏的術語，如 RAG、RAG 或 React 時，它們比你想象的要簡單得多。

Kicking things off at level one, large language models.

從第一層開始，大型語言模型。

Popular AI chatbots like chatGPT, Google Gemini, and Cloud are applications built on top of large language models, LLMs, and they're fantastic at generating and editing text.

流行的人工智能哈拉機器人，如 chatGPT、Google Gemini 和 Cloud，都是建立在大型語言模型（LLM）之上的應用程序，它們在生成和編輯文本方面非常出色。

下面是一個簡單的可視化例子。

You, the human, provides an input, and the LLM produces an output based on its training data.

人類提供輸入，LLM 根據訓練數據產生輸出。

For example, if I were to ask chatGPT to draft an email requesting a coffee chat, my prompt is the input, and the resulting email, that's way more polite than I would ever be in real life, is the output.

例如，如果我讓 chatGPT 起草一封請求喝咖啡的電子郵件，我的提示就是輸入，而結果就是輸出，這封郵件比我在現實生活中要禮貌得多。

But what if I ask chatGPT when my next coffee chat is?

但如果我問 chatGPT 下一次咖啡哈拉是什麼時候呢？

Even without seeing the response, both you and I know chatGPT is going to fail because it doesn't know that information.

即使沒有看到回覆，你我都知道 chatGPT 會失敗，因為它不知道這些資訊。

This highlights two key traits of large language models.

這突出了大型語言模型的兩個關鍵特徵。

First, despite being trained on vast amounts of data, they have limited knowledge of proprietary information, like our personal information or internal company data.

首先，儘管他們接受過海量數據方面的培訓，但對我們的個人信息或公司內部數據等專有資訊瞭解有限。

They wait for our prompt and then respond, right?

他們在等待我們的提示，然後做出迴應，對嗎？

Keep these two traits in mind moving forward.

在前進的道路上，請牢記這兩個特點。

進入第二層，人工智能工作流程。

What if I, a human, told the LLM, every time I ask about a personal event, perform a search query and fetch data from my Google calendar before providing a response?

如果我這個人告訴法律碩士，每次我詢問個人事件時，都要先進行搜索查詢，並從我的谷歌日曆中獲取數據，然後再提供答覆，會怎麼樣呢？

With this logic implemented, the next time I ask, when is my coffee chat with Elon Husky?

有了這個邏輯，下次我再問，什麼時候和埃隆-赫斯基喝咖啡哈拉？

I'll get the correct answer because the LLM will now first go into my Google calendar to find that information.

我將得到正確的答案，因為法律碩士現在會首先進入我的谷歌日曆查找該資訊。

What if my next follow-up question is, what will the weather be like that day?

如果我的下一個追問是：那天的天氣如何？

The LLM will now fail at answering the query because the path we told the LLM to follow is to always search my Google calendar, which does not have information about the weather.

現在，LLM 將無法回答該查詢，因為我們告訴 LLM 的路徑是始終搜索我的 Google 日曆，而該日曆並不包含天氣資訊。

This is a fundamental trait of AI workflows.

這是人工智能工作流程的基本特徵。

They can only follow predefined paths set by humans.

它們只能沿著人類預先設定的路徑前進。

And if you want to get technical, this path is also called the control logic.

如果你想深入瞭解，這條路徑也被稱為控制邏輯。

Pushing my example further, what if I added more steps into the workflow by allowing the LLM to access the weather via an API and then just for fun, use a text-to-audio model to speak the answer?

再進一步舉例說明，如果我在工作流程中增加更多步驟，讓 LLM 通過應用程序接口訪問天氣資訊，然後使用文本到音頻模型說出答案，會怎麼樣？

The weather forecast for seeing Elon Husky is sunny, with a chance of being a good boy.

天氣預報說，見到伊隆-哈斯基時天氣晴朗，有可能是個好孩子。

Here's the thing, no matter how many steps we add, this is still just an AI workflow.

問題是，無論我們添加多少步驟，這仍然只是一個人工智能工作流程。

Even if there were hundreds or thousands of steps, if a human is the decision maker, there is no AI agent involvement.

即使有成百上千個步驟，如果決策者是人類，也不會有人工智能代理的參與。

Pro tip, retrieval augmented generation, or RAG, is a fancy term that's thrown around a lot.

專業提示："檢索增強生成"（Retrieval augmented generation，簡稱 RAG）是一個常用的花哨術語。

In simple terms, RAG is a process that helps AI models look things up before they answer, like accessing my calendar or the weather service.

簡單來說，RAG 是一個幫助人工智能模型在回答問題前查找相關資訊的過程，比如訪問我的日曆或天氣服務。

Essentially, RAG is just a type of AI workflow.

從本質上講，RAG 只是一種人工智能工作流程。

By the way, I have a free AI toolkit that cuts through the noise and helps you master essential AI tools and workflows.

順便說一下，我有一個免費的人工智能工具包，它可以幫助你穿過噪音，掌握基本的人工智能工具和工作流程。

Following Helena Liu's amazing tutorial, I created a simple AI workflow using make.com.

按照 Helena Liu 的精彩教程，我使用 make.com 創建了一個簡單的人工智能工作流程。

Here, you can see that first, I'm using Google Sheets to do something.

在這裡，你可以看到，首先，我在使用 Google Sheets 做一些事情。

Specifically, I'm compiling links to news articles in a Google Sheet, and this is that Google Sheet.

具體來說，我正在用 Google Sheet 彙編新聞文章的鏈接，這就是那個 Google Sheet。

Second, I'm using Perplexity to summarize those news articles.

其次，我正在使用 Perplexity 來總結這些新聞報道。

Then, using Claude and using a prompt that I wrote, I'm asking Claude to draft a LinkedIn and Instagram post.

然後，利用克勞德和我寫的提示，我要求克勞德起草一篇 LinkedIn 和 Instagram 上的帖子。

Finally, I can schedule this to run automatically every day at 8 a.m.

最後，我可以將其安排在每天上午 8 點自動運行。

As you can see, this is an AI workflow because it follows a predefined path set by me.

如你所見，這是一個人工智能工作流程，因為它遵循的是我預先設定的路徑。

And finally, remember to run daily at 8 a.m.

最後，記得每天早上 8 點跑步。

One last thing, if I test this workflow and I don't like the final output of the LinkedIn post, for example, as you can see right here, it's not funny enough and I'm naturally hilarious, right?

最後一點，如果我測試了這個工作流程，但我不喜歡 LinkedIn 帖子的最終輸出結果，例如，你可以看到這裡，它不夠有趣，而我天生就很搞笑，對嗎？

I'd have to manually go back and rewrite the prompt for Claude, okay?

我得手動回去為克勞德重寫提示，好嗎？

And this trial and error iteration is currently being done by me, a human.

而這種試驗和錯誤迭代目前是由我這個人類完成的。

Continuing the make.com example, let's break down what I've been doing so far as the human decision maker.

繼續以 make.com 為例，讓我們分解一下我作為人類決策者迄今為止所做的工作。

With the goal of creating social media posts based off of news articles, I need to do two things.

為了根據新聞報道創建社交媒體帖子，我需要做兩件事。

First, reason or think about the best approach.

I need to first compile the news articles, then summarize them, then write the final posts.

我需要先整理新聞報道，然後進行總結，最後撰寫文章。

I need to find and link to those news articles in Google Sheets, use perplexity for real-time summarization, and then Claude for copywriting.

我需要在 Google Sheets 中找到這些新聞文章並與之鏈接，使用 perplexity 進行實時總結，然後使用 Claude 進行文案寫作。

So, and this is the most important sentence in this entire video, the one massive change that has to happen in order for this AI workflow to become an AI agent is for me, the human decision maker, to be replaced by an LLM.

所以，這是整段視頻中最重要的一句話，要讓人工智能工作流成為人工智能代理，必鬚髮生的一個巨大變化是，我這個人類決策者將被法律碩士所取代。

In other words, the AI agent must reason.

換句話說，人工智能代理必須進行推理。

What's the most efficient way to compile these news articles?

彙編這些新聞報道的最有效方法是什麼？

Should I copy and paste each article into a Word document?

我是否應該將每篇文章複製並粘貼到 Word 文檔中？

No, it's probably easier to compile links to those articles and then use another tool to fetch the data.

不，編譯這些文章的鏈接，然後使用另一種工具獲取數據可能更容易。

The AI agent must act, aka do things via tools.

人工智能代理必須採取行動，也就是通過工具做事。

Should I use Microsoft Word to compile links?

我應該使用 Microsoft Word 編輯鏈接嗎？

No, inserting links directly into rows is way more efficient.

不，直接在行中插入鏈接更有效。

No, the user has already connected their Google account with make.com, so Google Sheets is a better option.

不，用戶已經將其 Google 賬戶與 make.com 相連，是以 Google Sheets 是更好的選擇。

Pro tip, because of this, the most common configuration for AI agents is the React framework.

專業提示，正因為如此，人工智能代理最常用的配置是 React 框架。

All AI agents must reason and act, so react.

所有人工智能代理都必須進行推理和行動，是以必須做出反應。

Sounds simple once we break it down, right?

A third key trait of AI agents is their ability to iterate.

人工智能代理的第三個關鍵特徵是迭代能力。

Remember when I had to manually rewrite the prompt to make the LinkedIn post funnier?

還記得為了讓 LinkedIn 上的帖子更有趣，我不得不手動重寫提示語嗎？

I, the human, probably need to repeat this iterative process a few times to get something I'm happy with, right?

我，人類，可能需要重複這個迭代過程幾次，才能得到我滿意的東西，對嗎？

An AI agent will be able to do the same thing autonomously.

人工智能代理將能夠自主完成同樣的事情。

In our example, the AI agent would autonomously add in another LM to critique its own output.

在我們的例子中，人工智能代理將自主添加另一個 LM，對自己的輸出結果進行點評。

Okay, I've drafted V1 of a LinkedIn post.

好了，我已經起草了 LinkedIn 帖子的 V1。

Oh, I know, I'll add another step where an LM will critique the post based on LinkedIn best practices.

哦，我知道了，我會再加一個步驟，讓 LM 根據 LinkedIn 最佳實踐對帖子進行點評。

And let's repeat this until the best practices criteria are all met.

如此反覆，直到達到最佳實踐標準為止。

And after a few cycles of that, we have the final output.

經過幾個循環後，我們就得到了最終輸出結果。

That was a hypothetical example, so let's move on to a real world AI agent example.

這只是一個假設的例子，下面我們來看看現實世界中人工智能代理的例子。

Andrew Ng is a preeminent figure in AI, and he created this demo website that illustrates how an AI agent works.

吳恩達（Andrew Ng）是人工智能領域的傑出人物，他創建的這個演示網站展示了人工智能代理如何工作。

I'll link the full video down below, but when I search for a keyword like skier, enter, the AI vision agent in the background is first reasoning what a skier looks like.

我會在下面鏈接完整的視頻，但當我搜索滑雪者這樣的關鍵詞時，輸入，後臺的人工智能視覺代理會首先推理出滑雪者的樣子。

A person on skis going really fast in snow, for example, right?

比如說，一個人穿著滑雪板在雪地上飛馳，對嗎？

And then it's acting by looking at clips in video footage, trying to identify what it thinks a skier is, indexing that clip, and then returning that clip to us.

然後，它通過查看視頻錄像中的片段來採取行動，試圖識別它認為的滑雪者，為該片段建立索引，然後將該片段返回給我們。

Although this might not feel impressive, remember that an AI agent did all that instead of a human reviewing the footage beforehand, manually identifying the skier and adding tags like skier, mountain, ski, snow.

雖然這可能並不令人印象深刻，但請記住，這一切都是人工智能代理完成的，而不是人類事先查看錄像，手動識別滑雪者並添加滑雪者、山、滑雪、雪等標籤。

The programming is obviously a lot more technical and complicated than what we see in the front end, but that's the point of this demo, right?

編程顯然比我們在前端看到的技術性更強、更復雜，但這正是這個演示的意義所在，不是嗎？

The average user like myself wants a simple app that just works without me having to understand what's going on in the backend.

像我這樣的普通用戶需要的是一個簡單的應用程序，只需運行即可，無需瞭解後臺發生了什麼。

Speaking of examples, I'm also building my very own basic AI agent using NAN, so let me know in the comments what type of AI agent you'd like me to make a tutorial on next.

說到示例，我也在使用 NAN 構建我自己的基礎人工智能代理，所以請在評論中告訴我，你希望我接下來製作哪種類型的人工智能代理教程。

To wrap up, here's a simplified visualization of the three levels we covered today.

最後，下面是我們今天介紹的三個層次的簡化可視化示意圖。

Level one, we provide an input and the LLM responds with an output, easy.

第一層，我們提供輸入，LLM 響應輸出，很簡單。