Name: 大科技公司進入恐慌模式...DeepSeek R1是否戳破了人工智能保麗龍？ (Big Tech in panic mode... Did DeepSeek R1 just pop the AI bubble?)
Uploaded: 2025-02-03T05:06:56.000Z
Duration: 3 min 37 s

Last week, Chinese company DeepSeek shocked the world when they walked right over OpenAI's moat by releasing the open-source R1 model.

上週，中國公司 DeepSeek 發佈了開源 R1 模型，直接跨過了 OpenAI 的護城河，震驚了世界。

情態副詞

Not only does it beat their $200 O1 reasoning model, but it also beats CloudSonic and Gemini.

它不僅擊敗了售價 200 美元的 O1 推理模型，還擊敗了 CloudSonic 和 Gemini。

進階形容詞比較級句型

It wins on key benchmarks, but more importantly, it passes my vibe test, along with many others on the internet, as long as you don't ask it what happened on April 15th, 1989.

只要你不問它 1989 年 4 月 15 日發生了什麼，它就能通過我和互聯網上其他許多人的振動測試。

R1 even wins at math, which is crazy considering OpenAI had the answers in advance.

R1 甚至在數學上獲勝，考慮到 OpenAI 已經提前知道了答案，這簡直太瘋狂了。

Over the last few days, Normies have been learning about R1, and it went viral to become the number one app in America.

在過去的幾天裡，諾曼底人都在學習 R1，它以病毒式的傳播方式成為美國第一大應用程序。

This is a big problem for big tech, because in order for the grift to keep on grifting, artificial intelligence needs to be hard and take thousands of GPUs to pull off.

這對大型科技公司來說是個大問題，因為為了讓騙局繼續下去，人工智能必須非常困難，需要成千上萬的 GPU 才能完成。

But DeepSeek R1 cost less than $10 million, and was actually the side project of a hedge fund, where their only goal is to create long-term social value.

但 DeepSeek R1 的成本不到 1 千萬美元，而且實際上是一家對沖基金的副業，他們唯一的目標就是創造長期的社會價值。

Wall Street is terrified, and Nvidia's stock is in for a bloodbath today.

華爾街被嚇壞了，Nvidia 的股價今天將遭遇血洗。

Nvidia has been the primary beneficiary of the AI boom, but the whole bubble just got one-shotted by a brutal reality check.

英偉達一直是人工智能熱潮的主要受益者，但整個保麗龍剛剛被殘酷的現實一擊而破。

In today's video, we'll find out how DeepSeek just changed the course of history.

在今天的視頻中，我們將瞭解 DeepSeek 是如何改變歷史進程的。

It is January 27th, 2025, and you're watching The Code Report.

現在是 2025 年 1 月 27 日，您正在收看的是《密碼報告》。

Big tech is literally shaking right now, as Sam Altman coped with a dramatic tweet.

山姆-奧特曼（Sam Altman）在推特上發表了一條戲劇性的推文，大型科技公司現在簡直在顫抖。

A revolution can neither be made nor stopped.

革命既不能進行，也不能停止。

The only thing that can be done is for one of several of its children to give it direction by dint of victories.

唯一能做的，就是讓它的幾個孩子中的一個通過勝利為它指明方向。

I'm not sure if that's Napoleon Dynamite or Napoleon Bonaparte, but the funny thing is he damn near got ratioed by someone with an anime profile picture that said, it's not that deep, bro, just release a better model.

我不知道那是拿破崙-迪納摩還是拿破崙-波拿巴，但有趣的是，他差點就被一個有著動漫個人照片的人給比下去了，那人說："沒那麼深，兄弟，只是推出了一個更好的模型。

In addition to try to remain competitive with DeepSeek, OpenAI is also offering its O1 mini model to free users.

為了保持與 DeepSeek 的競爭力，OpenAI 還向免費用戶提供 O1 mini 機型。

They also released a new thing called Operator, which gives AI access to a browser, and allows it to click on buttons and fill out forms on your behalf, which is a game-changer, because that means you can have ChatGPT go attend your court-appointed DUI courses, or anything else you might have to do online.

他們還發布了一款名為 "操作員"（Operator）的新產品，讓人工智能可以訪問瀏覽器，並能代表你點擊按鈕和填寫表格，這改變了遊戲規則，因為這意味著你可以讓 ChatGPT 參加法院指定的酒駕課程，或其他任何你可能需要在線完成的事情。

In the future, humans will no longer need websites or UIs, only APIs that can interact with the AIs.

未來，人類不再需要網站或用戶界面，只需要能與人工智能互動的應用程序接口。

Operator is currently only available to people on the $200 plan, and overall it gets pretty mixed reviews.

目前只有 200 美元套餐的用戶才能使用 Operator，總體評價褒貶不一。

That's cool and all, but the Chinese are really crushing it right now.

這很酷，但中國人現在真的很厲害。

DeepSeek is one thing, but they also released this thing called Hanyuan 3D, which can generate 3D meshes and textures, and that's huge for people like me, who have spent years failing to get good at Blender.

DeepSeek 是一回事，但他們還發布了一款名為 Hanyuan 3D 的軟件，可以生成 3D 網格和紋理，這對於像我這樣多年來都無法熟練使用 Blender 的人來說意義重大。

But the crazy thing about DeepSeek is that it was just a side project that cost very little money to produce.

但 DeepSeek 的瘋狂之處在於，它只是一個製作成本極低的副業。

Even Marc Andreessen called it one of the most amazing and impressive breakthroughs he's ever seen, a profound gift to the world.

就連馬克-安德森（Marc Andreessen）也稱這是他所見過的最驚人、最令人印象深刻的突破之一，是獻給世界的一份厚禮。

And because of that, the stock market is panicking, and just wiped out nearly a trillion dollars in value, with Nvidia being the biggest loser, followed by other chip companies like Taiwan Semiconductor and Broadcom.

正因為如此，股市出現恐慌，剛剛抹去了近萬億美元的價值，Nvidia 是最大的輸家，其次是臺灣半導體和博通等其他芯片公司。

You see, Nvidia's had basically a monopoly on AI training for the last decade, because they have better Linux drivers than AMD, and all the machine learning algorithms are optimized for libraries like CUDA, which is proprietary and can only run on Nvidia GPUs.

在過去十年中，Nvidia 基本上壟斷了人工智能培訓領域，因為他們擁有比 AMD 更好的 Linux 驅動程序，而且所有機器學習算法都針對 CUDA 等庫進行了優化，而 CUDA 是專有的，只能在 Nvidia GPU 上運行。

And they have the infrastructure to connect all these GPUs together at scale.

他們還擁有將所有這些 GPU 大規模連接在一起的基礎設施。

As AI models get bigger and more popular, so do the profits of Nvidia.

隨著人工智能模型越來越大、越來越流行，Nvidia 的利潤也越來越高。

But someone got DeepSeek running on a few Apple M2 Ultras, and that's the 680 billion parameter top-of-the-line model.

但有人在幾臺蘋果 M2 Ultras 上運行了 DeepSeek，那是參數為 6800 億的頂級機型。

And that's terrifying if you're a tech investor pouring billions of dollars into AI, because if the best state-of-the-art model can operate on relatively cheap consumer hardware, and it's free and open source and developed in China, how are we going to make any mother-effin' money?

如果你是一位正在為人工智能投入數十億美元的科技投資者，這就太可怕了，因為如果最先進的模型可以在相對廉價的消費硬件上運行，而且它是免費開源的，是在中國開發的，那我們還怎麼賺錢？

People are calling this the Sputnik moment.

人們稱之為 "斯普特尼克時刻"。

When Russia launched the first artificial satellite into space in 1957.

1957 年，俄羅斯向太空發射了第一顆人造衛星。

The United States was shocked at the time, and it took over a decade to make a comeback, when America made its pinnacle achievement of filming a man on the moon in Nevada.

當時美國震驚了，過了十幾年才捲土重來，美國在內華達州取得了拍攝人類登月的巔峰成就。

That's a joke, of course, and even the smartest technology in the world, DeepSeek, believes in the moon landings.

這當然是個笑話，就連世界上最聰明的技術 DeepSeek 也相信登月。

We had a good run on this AI hype train, but now it's time to move on.

我們在人工智能的炒作列車上跑了一圈，但現在是時候繼續前進了。

I'm looking forward to brain chips, quantum computers, artificial meat printers, nanorobots, digital resurrection, and all kinds of other tech that will make the world a better place.

我期待著大腦芯片、量子計算機、人造肉打印機、納米機器人、數字復活以及其他各種能讓世界變得更美好的技術。

 Thanks for watching, and I will see you in the next one.

感謝您的收看，我們下期再見。