Artificial intelligence. - Job.

人工智慧。- 工作機會。

Artificial intelligence.

人工智慧。

You've heard the AI warnings.

你已經聽到人工智慧的警告了。

Every technological revolution affects the job market.

每一次科技革命都會影響就業市場。

AI could cost some humans their jobs in the near future.

在不久的將來，人工智慧可能會讓一些人類失去工作。

But AI is also creating some new jobs, specifically one.

但是，人工智慧也創造了一些新的工作職位，特別是其中之一。

So I decided to apply.

於是我決定應徵。

Job title: prompt engineer.

職位名稱：AI 溝通師。

Responsibilities: writing text commands or prompts to get the best answers out of a generative AI large language model like ChatGPT.

職責：編寫文字命令或提示，以從 ChatGPT 等生成式 AI 大語言模型中獲得最佳答案。

Pay: up to $250,000 a year.

薪酬：最高年薪 25 萬美元。

To figure out exactly what a prompt engineer does, I decided to apply to be one at Hebbia, a New York-based AI startup.

為了弄清 AI 溝通師到底是做什麼的，我決定應徵成為 Hebbia 的一員，這是一家位於紐約的 AI 新創公司。

Wish me luck.

祝我好運吧。

- Hi - Swetha. Good to meet you.

- 你好。 - Swetha，很高興見到你。

- Nice to meet you. - Welcome to the office.

- 幸會。- 歡迎來到辦公室。

I have wanted to be a prompt engineer as long as I can remember.

從我有記憶以來，我一直想成為一名 AI 溝通師。

All right, fine, the company was in on this little experiment.

好吧，好吧，公司參與了這個小實驗。

It's really interesting that you say that because the role actually didn't exist until maybe a year ago.

你這麼說真的很有趣，因為這個工作實際上直到大約一年前才出現。

Yes, prompt engineering is a totally new type of job because well, large language models or LLMs, like ChatGPT only hit the scene a year ago.

是的，AI 溝通師是一種全新的工作類型，因為像 ChatGPT 這樣的大型語言模型或 LLM 一年前才出現。

And if you've used ChatGPT, prompt engineering is actually pretty easy to understand.

如果你使用過 ChatGPT，AI 溝通師的工作其實很容易理解。

You type in your prompt, instructing whatever you'd like the large language model to do.

你輸入你的指令時，指示你希望大型語言模型執行的任何操作。

The model then returns an answer, but sometimes that answer is unwieldy or not what you wanted.

然後模型返回一個答案，但有時該答案很笨拙或不是你想要的。

The better the input, the better the output.

指令下的越好，輸出的答案就越好。

For really one of the first times ever,

這真是有史以來的第一次，

we've been able to expose a pretty complicated piece of technology to someone that doesn't know how to code,

我們已經能夠向不知道如何寫程式的人展示一項相當複雜的技術，

to someone that knows how to talk to these models in a highly instructive, highly clear way.

向知道如何以高度指導性、高度清晰的方式與這些模型對話的人展示。

And that person is the prompt engineer.

這個人就是 AI 溝通師。

So really it's a translator.

這職業其實就是一個翻譯。

That's exactly what it is.

正是如此。

You're translating between the machine and the human.

你在機器和人類之間進行翻譯。

Seems like something I could be pretty good at.

看來我應該還挺在行的。

I think so too.

我也這麼認為。

But what's a good prompt and how do you learn to write them?

但什麼是好的指令以及該如何學習寫指令呢？

Well, I went back to school and by school, I mean a Coursera class.

我回到學校，繼續學習，這裡的學校指的是 Coursera 課程。

Look at that. 100 and 100.

看看這個。100 和 100。

Prompts aren't just a one-off that you give to the large language model.

指令不只是一次性的的語言模型。

In his six-week prompt engineering for ChatGPT class,

在為期六週的 ChatGPT 指令工程課程中、

Dr. Jules White, a professor of computer science at Vanderbilt University teaches you how to structure sentences to get the desired output from the bot.

范德比爾特大學電腦科學教授 Jules White 博士教你如何構建句子以從機器人獲得所需的輸出。

You seem bored by this, Browser.

瀏覽器，你似乎對此感到厭煩。

There's the persona or role prompt.

有人物或角色指令。

For instance, act as an assistant helping a financial analyst.

例如，擔任財務分析師的助理。

You're instructing the LLM to take on a persona.

你正在指示 LLM 扮演一個角色。

There's the new information prompt.

這是新的資訊指令。

This is where you feed the model new data it might not know about.

你可以在此處向模型提供它可能不知道的新數據。

Say, here's the transcript from the Microsoft earnings call. Please summarize the mentions of Bing, and you get a summary.

比如說，這是微軟財報電話會議的文字記錄。 請總結 Bing 的提及，你會得到一個摘要。

Hebbia's Matrix app actually does all of this.

Hebbia 的 Matrix 應用程式就能做到這些。

It allows financial institutions and other businesses to upload or access huge datasets, filings, earnings transcripts, calls.

它允許金融機構和其他企業上傳或存取大量資料集、文件、收益記錄、電話。

Then it uses AI to find answers so people don't have to manually go through it all.

然後它使用人工智慧來尋找答案，這樣人們就不必手動完成這一切。

And that's what Swetha gave me as a prompt engineering assignment.

這就是 Swetha 給我的 AI 溝通任務。

So in this case, let's say that you are doing an analysis on earnings calls for a public company.

因此，在這種情況下，假設你正在分析一家上市公司的收益電話會議。

I did a pretty good job at writing those prompts, but there was another part of the job I wasn't prepared for: coding.

我在編寫這些指令方面做得相當不錯，但工作的另一部分我還沒準備好：寫程式。

In some cases, a prompt engineer may have to code the queries into the tool itself so Hebbia's clients can use them with a single tap.

在某些情況下，AI 溝通師可能必須將問題寫進工具本身中，以便 Hebbia 的客戶只需點擊一下即可使用它們。

It was about that time that I knew this probably wasn't my next job.

就在那時，我知道這可能是不是我的下一份工作。

Plus, I was a little worried about how long this job might be a job, which I decided to bring up in my interview with the CEO.

另外，我有點擔心這份工作的前景會有多久，我決定在與執行長的訪談中提出這個問題。

Joanna. Nice to meet you.

Joanna，很高興見到你。

As these models get better, is prompt engineering gonna be necessary?

隨著這些模型變得越來越好，AI 溝通師會成為必要職業嗎？

At the end of the day, you'll still need people to tell them explicit instructions.

說到底，這些工具仍然需要人類來告訴他們明確的指令。

Even in programming where machines are perfect at understanding code, there's still a job for people to go back and forth.

即使在機器能夠完美理解程式碼的程式設計中，仍然需要人類來回確認。

So what did I learn today?

我今天學到了什麼？

Well, getting a job as a prompt engineer is harder than I thought.

找到一份 AI 溝通師的工作比我想像的還要難。

Still, anyone who wants to successfully work with AI tools and elicit the best answers from them really should learn some of these prompt writing skills,

儘管如此，任何想要成功使用人工智慧工具並從中得出最佳答案的人確實應該學習一些寫指令的技巧，

even if your next job title isn't officially prompt engineer.

即使你的下一個職位不是正式的 AI 溝通師。

Joanna, enjoyed meeting you too.

Joanna，很高興認識你。

After some deliberation, we will not be continuing the hiring process here.

經過一番深思熟慮，我們將不再繼續你的招聘程序。

Wishing you the best in your search.

祝你在尋找工作的過程中一切順利。

Yeah, rough.

是啊，真糟糕。