Name: 十足的納粹埃隆-馬斯克的人工智能哈拉機器人開始自稱 "機械希特勒 (‘Full Nazi’: Elon Musk's AI chatbot started calling itself 'MechaHitler')
Uploaded: 2025-07-16T14:12:57.000Z
Duration: 6 min 10 s

程度副詞

Among all the other things happening at X today, the company's CEO, a woman named Linda Yaccarino, announced she's leaving after just two years on the job.

在 X 公司今天發生的所有其他事情中，該公司的首席執行官，一位名叫琳達-雅卡里諾（Linda Yaccarino）的女士，宣佈她將在任職兩年後離職。

In a lengthy post, Yaccarino wrote, quote, When Elon Musk and I first spoke of his vision for X, I knew it would be the opportunity of a lifetime to carry out the extraordinary mission of this company.

亞卡里諾在一篇長文中寫道："當埃隆-馬斯克和我第一次談到他對 X 公司的願景時，我就知道這將是我一生中難得的機會，來完成這家公司非凡的使命。

情態副詞

I'm immensely grateful to him for entrusting me with the responsibility of protecting free speech, turning the company around and transforming X into the everything app.

我非常感謝他將保護言論自由、扭轉公司局面以及將 X 轉變為無所不能的應用程序的重任交給我。

Elon Musk simply responded, quote, Thank you for your contributions.

埃隆-馬斯克（Elon Musk）簡單地迴應道："謝謝你們的貢獻。

After Musk bought the site, Yaccarino was brought in to be the adult in the room, I guess, but she may have gotten more than she bargained for with her boss's rants, a platform that seems to court right-wing extremists, and a high-profile AI chatbot that refers to itself as Mecca Hitler.

馬斯克收購該網站後，雅卡里諾被請來充當 "成年人"，我猜，但她得到的可能比她想要的還要多：她老闆的咆哮、一個似乎在討好右翼極端分子的平臺，以及一個自稱為 "麥加希特勒 "的高調人工智能哈拉機器人。

Brian Barrett is the executive editor of News at Wired, which has done extensive coverage of Grok, Elon Musk, and he joins me now.

布萊恩-巴雷特（Brian Barrett）是《連線》新聞頻道的執行主編，《連線》曾對格魯克和埃隆-馬斯克進行過廣泛報道，他現在加入了我的談話。

Brian, let me first start on the Yaccarino news and the question of whether we know if her departure was at all related to the extraordinary and appalling meltdown of the Grok AI system yesterday.

布萊恩，首先讓我談談雅卡里諾的新聞，以及我們是否知道她的離職是否與昨天格羅克人工智能系統令人震驚的異常崩潰有關。

You know, I'll say first that we don't know.

你知道，我首先要說我們不知道。

You could certainly see how it might be the last straw, but I will say there were plenty of other straws before this that could have been the last one.

你當然可以理解這可能是最後一根稻草，但我要說的是，在此之前還有很多其他稻草可能是最後一根稻草。

If you're not leaving over the white genocide incident that you mentioned that happened in May, you know, this is worse, but it's all pretty bad.

如果你還沒有從你提到的五月份發生的白人種族滅絕事件中解脫出來，你知道，這次的情況更糟，但都很糟糕。

I think there are some other factors to consider here.

我認為還有其他一些因素需要考慮。

It got absorbed by the AI company that Elon Musk also owns, so her role is diminished there.

它被埃隆-馬斯克也擁有的人工智能公司兼併了，所以她在那裡的作用被削弱了。

And also, she came in as someone who could make nice with advertisers, bring them back to the platform.

同時，她也是一個能與廣告商友好相處的人，能把他們帶回平臺。

Now that X is evolving into, A, an everything app with payments and who knows what else, and B, an AI forward Mecca Hitler app, it's less clear what her role would be.

現在，X 正在演變成：A，一個包含支付和其他功能的萬能應用；B，一個人工智能前沿的麥加希特勒應用，她的角色就不那麼明確了。

So it makes sense that she would like, it makes sense that she would head out now, but I don't know that you could say this is the one thing where she finally said enough is enough.

是以，她現在想離開是有道理的，但我不知道你是否可以說這是她終於說 "夠了 "的一件事。

Okay, let's talk about what happened with Grok.

好吧，我們來談談格羅克的事。

I was sort of horrified and fascinated in equal measure.

I mean, it's clear that, I mean, there's a thing that happens on X all the time, which is there'll be something and then Grok will fact check it and seem to be kind of a liberal or correct right wing misinformation.

我的意思是，很明顯，我的意思是，在X上經常會發生這樣的事情，就是會有一些事情，然後格羅克會對其進行事實核查，似乎是一種自由派或糾正右翼的錯誤信息。

And then all the right wing people on the app get so mad at it.

然後，應用程序上的所有右翼人士都對此感到非常生氣。

So do we actually, A, do we know why this happened?

那麼，我們是否知道，A，為什麼會發生這種情況？

And B, how, why it took them so long to put a stop to it?

B，怎麼會，為什麼他們花了這麼長時間才制止？

Well, in terms of why it happened, I think you can look at, you mentioned before, they added these system prompts that said, don't shy away from politically incorrect material.

至於為什麼會這樣，我想你可以看看，你之前提到過，他們添加了這些系統提示，說不要回避政治不正確的材料。

And there was another one there, too, that said, don't necessarily trust the media, media sources.

還有一個人也說，不一定要相信媒體，媒體的消息來源。

So that combined was a factor, certainly.

But then also you got to remember that these AI models, when your knowledge base, when your training data comprises the entirety of the Internet, there's a lot of really bad stuff on the Internet, which we all know.

但你也要記住，這些人工智能模型的知識庫和訓練數據是由整個互聯網組成的，互聯網上有很多非常糟糕的東西，這是我們都知道的。

So when you don't put in the right guardrails or when you remove guardrails, it's garbage in, garbage out.

是以，如果不安裝正確的護欄，或者拆除護欄，就會造成 "垃圾進，垃圾出"。

I think any of these AI models, if you let them run rampant, that's probably what they'll do.

我認為，任何一種人工智能模型，如果你讓它們橫衝直撞，它們可能就會這麼做。

That's why companies spend so much money trying to prevent this from happening.

這就是為什麼公司要花費大量資金來防止這種情況發生。

I mean, to me, the part that was fascinating, right, is to the degree to which, you know, the system prompt was don't worry about things that are politically incorrect.

我的意思是，對我來說，吸引人的部分是，你知道，系統提示的程度是，不要擔心政治上不正確的事情。

And then you train this LLM on all the content, the Internet that it just took that prompt.

然後，你就所有的內容和互聯網對這位法學碩士進行培訓，這就是它的提示。

That like the sort of neural network mapping of the conceptual deep meaning of don't be politically incorrect is go full Hitler at the level of the of the model itself.

就像神經網絡映射 "不要政治不正確 "這一概念的深層含義一樣，在模型本身的層面上，希特勒已經完全掌握了這一概念。

We want to be a little bit careful because we don't know for sure.

我們要小心一點，因為我們還不確定。

Like there may be stuff going on behind the scenes.

就像幕後可能會發生一些事情一樣。

There may be some other tweaks that are made, but fundamentally, yes, I think it's at least speaks to the ethos of, hey, we're going to let you run a little bit wild.

可能還會有一些其他的調整，但從根本上說，是的，我認為這至少體現了一種精神，那就是：嘿，我們要讓你有點野性。

You're going to take several miles in this case.

在這種情況下，你需要走幾英里路。

What what's what is the success level of X?

I feel like it was clear it was losing a ton of revenue.

我覺得它顯然損失了一大筆收入。

When Trump got elected, there was a sort of sense that like the administration and X would successfully sort of bully advertisers to come back to the platform.

特朗普當選後，有一種感覺是，政府和 X 會成功地欺負廣告商，讓他們回到平臺上來。

There was even reporting about, you know, administration officials doing that FTC investigation into possible conspiracy to obstruct trade.

甚至有報道稱，你知道，政府官員正在進行聯邦貿易委員會的調查，調查可能存在的共謀妨礙貿易的行為。

You know, I think when you talk about the success level of X, it's important not to get too caught up in dollars and cents.

你知道，我認為當你談論 X 的成功水準時，重要的是不要太糾結於美元和美分。

I think it's it's clearly lost a ton of users.

我認為它顯然失去了大量用戶。

It's clearly not a happy place for advertisers.

對於廣告商來說，這顯然不是一個令人愉快的地方。

I thought it was interesting that Linda Carino said that she had done so much to restore advertiser trust.

琳達-卡里諾說她為恢復廣告商的信任做了很多工作，我覺得這很有意思。

But I think a success for X is really how much can advance Elon Musk's agenda?

但我認為，X 的成功真正在於能在多大程度上推進埃隆-馬斯克的議程？

How much can it be his personal playground?

它有多大可能成為他的個人遊樂場？

How much can you use it as a lever for power?

你能在多大程度上利用它作為權力的槓桿？

And I think clearly it's done that in space.

我認為在太空中，它顯然做到了這一點。

And so I, you know, monetarily probably lost, but he's got 200 billion dollars.

所以我，你知道，在金錢上可能輸了，但他有 2000 億美元。

He doesn't care that much, I would imagine.

And it's still a really great place for him to hang out with his like minded buddies.

而且，這裡還是他和志同道合的夥伴們一起玩耍的好地方。

And his personal AI, Brian Barrett at Wired, which I have to say has been doing some of the finest reporting in this Trump administration.

他的個人助理、《連線》的布萊恩-巴雷特（Brian Barrett），我不得不說，他在特朗普政府中的一些報道非常出色。

I read you guys constantly on a subscriber and you're doing amazing work.

我經常在訂閱器上閱讀你們的文章，你們的工作非常出色。

So thank you for all the work you guys are doing out there.

是以，感謝你們在外面所做的一切工作。