48
0Episode: Mercor: AI Now Eats Dead Startups' Slack Chats
Duration: approximately 7 minutes
Level: B1 (Intermediate)
---
[Mike]: Welcome back to "Learn English with Podcasts"! Sarah, last week we found out that AI companies are buying old books. Well, today, they are hungry for something much more private.
zh:欢迎回到"Learn English with Podcasts"！Sarah，上周我们发现AI公司正在购买旧书。而今天，它们想吃的是私密得多的东西。
[Sarah]: More private than a bookshelf full of old novels? Now I am a little worried. What is it this time?
zh:比满满一架旧小说还私密？现在我开始有点担心了。这次又是什么？
[Mike]: Company chat records. The messages that people send to each other at work. An AI data company called Mercor is now buying these from startups that fail or get bought out.
zh:公司聊天记录。就是人们在工作中互相发送的消息。一家名叫Mercor的AI数据公司，现在正从倒闭或被收购的创业公司手里买这些东西。
[Sarah]: Mercor? I have heard that name somewhere. Are they the company behind OpenAI?
zh:Mercor？我在哪儿听过这个名字。他们是OpenAI背后的公司吗？
[Mike]: Not exactly. Mercor does not build AI models. It collects the data that AI companies need. People call it the "data factory" behind OpenAI and Anthropic.
zh:不完全是。Mercor不做AI模型。它收集AI公司需要的数据。人们把它叫做OpenAI和Anthropic背后的"数据工厂"。
[Sarah]: So Mercor is like a chef who buys ingredients for other restaurants. And the new ingredient is... other people's work chats?
zh:所以Mercor就像一位给其他餐厅买食材的厨师。而新食材是……别人的工作聊天记录？
[Mike]: Exactly. When a startup closes or gets bought, all its internal data suddenly has no owner. Slack messages, Google Drive files, emails, even Zoom meeting recordings. And Mercor wants to buy them.
zh:没错。当一家创业公司倒闭或被收购，它所有的内部数据突然就没了主人。Slack消息、Google Drive文件、电子邮件，甚至Zoom会议录音。而Mercor想买下这些。
[Sarah]: Wait, someone buys a whole company's chat history? Who would even sell that?
zh:等等，有人会买整个公司的聊天记录？谁会卖这种东西啊？
[Mike]: Here is a real story. Warmly is an AI company that makes sales agents. At the end of June, it announced that HubSpot would buy it. Just eight days later, the CEO, Max Greenwald, received an email from Mercor.
zh:给你讲个真实的故事。Warmly是一家做销售智能体的AI公司。6月底，它宣布HubSpot要收购它。仅仅8天后，CEO马克斯·格林沃尔德就收到了Mercor的邮件。
[Sarah]: Eight days? That is fast. What did the email say?
zh:才8天？这速度真快。邮件里说了什么？
[Mike]: Mercor wanted to buy or license their Slack chats, GitHub records, project files, and meeting transcripts. The offer was up to three hundred thousand dollars. Greenwald said no. And when more offers came from other data companies, he said no to all of them.
zh:Mercor想购买或授权他们的Slack聊天记录、GitHub记录、项目文件和会议转录。报价最高达到30万美元。格林沃尔德拒绝了。之后其他数据公司又发来更多报价，他也全都拒绝了。
[Sarah]: Three hundred thousand dollars for old work messages, and he refused? Good for him. But why would anyone pay so much money?
zh:为一些旧的工作消息出30万美元，而他拒绝了？他做得对。但为什么会有人愿意付这么多钱？
[Mike]: Let me explain with another company, called Protege. Its CEO, Bobby Samuels, says AI labs have basically scraped the whole internet already. Books, websites, code, all eaten. Now they need something new.
zh:我用另一家公司Protege来解释。它的CEO鲍比·塞缪尔斯说，AI实验室基本上已经把整个互联网刮干净了。书、网站、代码，全都吃完了。现在它们需要新的东西。
[Sarah]: Something that is not on the internet?
zh:网上没有的东西？
[Mike]: Real human interactions. How people actually do their jobs. Protege helps companies decide if their data is worth selling. This year, they have handled over one hundred million dollars in data deals. Last year? About thirty million.
zh:真实的人际互动。人们到底是怎么完成工作的。Protege帮公司判断他们的数据值不值得卖。今年，他们已经经手了超过1亿美元的数据交易。去年呢？大约3000万。
[Sarah]: Wow, more than three times. So work chats are becoming very valuable.
zh:哇，三倍还多。所以工作聊天记录正变得非常值钱。
[Mike]: And here is the reason. AI is changing from a chatbot into a digital worker. A chatbot only needs to answer questions. But a digital worker needs to know how humans actually get things done.
zh:原因就在这里。AI正在从聊天机器人变成"数字员工"。聊天机器人只需要回答问题。但数字员工需要知道人类到底是怎么把事情做成的。
[Sarah]: Can you give me an example? Something I can actually picture in my head?
zh:你能举个例子吗？一个我能真正在脑子里看见的例子？
[Mike]: Sure. An employee writes an IT ticket: "I cannot log in." In the Slack chat, engineers discuss what might be wrong. An email has the final fix. And GitHub shows the code change.
zh:当然。一名员工提交了IT工单："我登录不上了。"在Slack聊天里，工程师们讨论可能是什么问题。邮件里有最终修复方案。而GitHub上记录了代码改动。
[Sarah]: To us, that is just a normal Tuesday at the office.
zh:对我们来说，那只是办公室里普通的一天。
[Mike]: To an AI agent, it is a complete work tutorial. Spot the problem, discuss the cause, write the fix, step by step. That is exactly what AI agents need for training.
zh:对AI智能体来说，那是一整套完整的工作教程。发现问题、讨论原因、写下修复方案，一步一步。这正是AI智能体训练最需要的东西。
[Sarah]: So when I ask my AI assistant to "log in and fix the problem", it already knows the steps... because it read someone's chat history?
zh:所以当我让我的AI助手"登录并修复问题"时，它已经知道步骤了……因为它读过别人的聊天记录？
[Mike]: That is the plan. There is just one big problem, privacy. Slack chats may contain customer information. Emails have employee names. Meetings might reveal business secrets. And healthcare data can include patient privacy.
zh:计划就是这样。但有一个大问题：隐私。Slack聊天里可能有客户信息。邮件里有员工姓名。会议中可能泄露商业机密。医疗数据甚至涉及患者隐私。
[Sarah]: So they cannot just sell it all. The data has to be cleaned first.
zh:所以不能直接卖。数据得先清理。
[Mike]: Anonymized. Names and private details are removed before the data can be sold. Only then can it feed an AI.
zh:匿名化处理。名字和隐私细节会被删掉，然后数据才能出售。只有这样，它才能喂给AI。
[Sarah]: Hmm. So a startup closes down. Its last messages are cleaned, packaged, and sold. And a future AI reads them to learn how to do a human job.
zh:嗯。所以一家创业公司倒闭了。它最后的消息被清理、打包、出售。然后一个未来的AI读它们，学习怎么干人类的活。
[Mike]: Exactly. Yesterday, AI was learning from old books. Today, it is learning from the small things we type at work. The quick messages, the questions, the fixes.
zh:没错。昨天，AI还在学旧书。今天，它学的是我们工作时打出的那些小东西。快速的留言、问题、还有修复方案。
[Sarah]: Oh no. So the real training data was never the books or the internet. It was us, all along. The moment I type "please fix this", I might be writing a page in an AI's work manual.
zh:天哪。所以真正的训练数据从来不是书本或互联网。一直都是我们。从我打出"请修复这个"那一刻起，我可能就在给AI写一页工作手册。
[Mike]: And that is why I am always polite in my chat messages. You never know who... or what... might read them later.
zh:所以我发聊天消息时总是很有礼貌。你永远不知道以后会是谁……或者是什么……读到它们。
[Sarah]: Ha! Very good point. Thanks for listening to "Learn English with Podcasts". See you next time!
zh:哈！说得太对了。感谢收听"Learn English with Podcasts"。下次见！
[Mike]: And remember, the next time your AI assistant finishes your work, it might be your own messages teaching it. Bye!
zh:还有记住，下次你的AI助手替你完成工作时，教它的可能正是你自己发的消息。再见！
