Name: 像英偉達（Nvidia）這樣的美國公司備受關注。但中國在人工智能領域做了些什麼？ (U.S. firms like Nvidia get the attention. But what's China doing in AI?)
Uploaded: 2025-10-28T17:56:28.000Z
Duration: 30 min 20 s

Welcome to beyond the Valley, a CNBC tech podcast with Arjun Karpil and Tom Chitty.

歡迎來到「科技谷之外」，CNBC 的科技Podcast，由 Arjun Karpil 和 Tom Chitty 主持。

形容詞最高級

For a few days in June, tech firm Nvidia became the most valuable company in the world.

在六月份的幾天裡，科技公司 Nvidia 成為了全球最有價值的公司。

動名詞

Investors see the American chip maker as a dominant player in artificial intelligence, with its hardware key to training huge AI models, including ChatGPT.

投資者認為這家美國晶片製造商是人工智慧領域的主導者，其硬體是訓練包括 ChatGPT 在內的龐大 AI 模型關鍵。

Meanwhile, US tech giants from Google to OpenAI are of part forging ahead with their own innovative AI models.

同時，從 Google 到 OpenAI 的美國科技巨頭們，也正在積極開發自家的創新 AI 模型。

But this hardware and software is not available to all.

但這些硬體和軟體並非人人都能取得。

Beijing, faced with US export restrictions, is seeking a homegrown alternative to American AI technology.

北京方面，面對美國的出口限制，正在尋求本土的替代方案來取代美國的 AI 技術。

And can China overcome the many challenges it faces to keep pace with its competitors?

中國能否克服面臨的諸多挑戰，跟上競爭對手的步伐？

Beyond the Valley no guests this week, but

「科技谷之外」本週沒有來賓，但是

Arjun, I know that you were really keen to tackle this, this particular topic.

Arjun，我知道你非常熱衷於探討這個，這個特別的主題。

You've obviously been reporting on it quite a lot over the last week or so.

你最近一週以來，顯然對這件事做了不少報導。

Is it really sort of piqued your interest?

Do you think this is a story that's just going to keep running and running?

你認為這個故事會一直持續下去嗎？

I've been doing a lot of research to understand what was going on in China at the moment.

我做了很多研究，想了解中國目前的情況。

We've been talking so much and hearing so much about companies like OpenAI, who you mentioned there in the open, the company behind ChatGPT, Google, Microsoft, Apple Intelligence and their AI features.

我們聊了好多，也聽了好多關於 OpenAI、你剛剛提到的 ChatGPT 背後的公司，還有 Google、Microsoft、Apple Intelligence 和他們的 AI 功能。

Yeah, particularly with it, with some serious challenges that they face in developing AI technology.

是的，特別是他們在開發 AI 技術時面臨一些嚴峻的挑戰。

But before we get into it, let's do quickly stat of the week.

但在我們深入探討之前，先來快速看看本週的統計數據。

Do you think you make it easier when it's just me or harder?

你覺得只有我一個人的時候，事情會比較容易還是比較難？

No, I think it's easier when it's just you because you've got less pressure from a competitor, which is you usually our super smart guest.

不，我覺得只有你一個人的時候比較容易，因為你沒有來自競爭對手的壓力，而競爭對手通常是我們非常聰明的嘉賓。

So for our listeners, first, let's establish, when we talk about AI models, you refer to them as foundational models.

所以，對於我們的聽眾，首先，我們先確立一下，當我們講到 AI models 的時候，你指的是 foundational models。

So effectively it's this sort of underpinning of an AI application trained on huge amounts of different data and they're sort of capable of a range of different tasks.

所以，基本上它就是一個 AI 應用程式的基礎，透過大量的不同數據訓練而成，而且它們能夠執行各種不同的任務。

You know, it could be something like text generation or audio generation.

So they're trained on vast amounts of data and one kind of foundational model is a large language model or LLM, which we've heard a lot about.

所以它們是透過海量數據訓練的，而一種 foundational model 就是大型語言模型，也就是 LLM，我們常聽到這個詞。

Again, these underpin things like ChatGPT, they're trained on huge amounts of text data, which allows them.

同樣地，這些是 ChatGPT 等的基礎，它們透過大量的文字數據訓練，這讓它們能夠。

For example, on ChatGPT, if you input text, you get a generative text response.

舉例來說，在 ChatGPT 上，如果你輸入文字，你會得到一個生成式的文字回應。

Similar now with images and audio and even video to some extent.

現在圖像、音訊，甚至某種程度上的影片也是類似的。

You can effectively see them as the model, the blueprint which underpins these applications that we use with AI.

你可以把它們看作是我們用 AI 使用的這些應用程式的基礎模型、藍圖。

And the role of chips in there give us a little bit more depth on that.

那晶片在其中的角色，能讓我們更深入瞭解。

So the, the models need to be trained with all of this data.

所以，這些模型需要用所有這些數據來訓練。

And this requires huge amounts of processing power and to get that processing power that comes from the chips.

這需要龐大的處理能力，而要獲得這些處理能力就來自晶片。

And these chips are often in the data centers, these huge literal warehouses of servers.

而這些晶片通常都在資料中心，也就是那些巨大的、真正的伺服器倉庫裡。

And the chips are all inside there effectively.

And they're running these, these huge processes to train up this AI model so that it has the ability to respond to queries, etc.

它們正在運行這些龐大的處理程序來訓練這些 AI 模型，讓它們能夠回應查詢等等。

And just to add to that, we talk about chips and there's a specific type of chip which is sort of hit the headlines and that's the graphics processing unit or gpu, the likes of which Nvidia designs.

再補充一點，我們談到晶片時，有一種特定的晶片最近備受矚目，那就是繪圖處理器，也就是 GPU，Nvidia 就設計了這類產品。

And again, you mentioned Nvidia at the top.

再次提到，您一開始就說了 Nvidia。

We'll talk a lot about more about that company, those kind of chips and you know, what China has on offer there.

我們會多談談這家公司、這類晶片，以及中國在那方面提供的東西。

Nvidia very much seen as the leading player in that space.

Nvidia 在這方面無疑是領先的廠商。

好的，我們聽了很多關於 OpenAI 的事。

I'm sure our listeners have heard of OpenAI, the various iterations of ChatGPT, GPT4, I think.

我確定我們的聽眾一定聽過 OpenAI，還有像是 ChatGPT、GPT4 等各種版本。

Where is the latest version currently, and other large language models that other companies are developing as well.

目前最新的版本在哪裡，還有其他公司也在開發的大型語言模型呢？

So can you tell us about the leading models in China?

所以，您能跟我們談談中國的領先模型嗎？

Because I'm sure our listeners have heard less about that and a little bit about what are they, what, what purpose do they serve and, and

因為我確定我們的聽眾對這方面比較不熟悉，想知道它們是什麼、有什麼用途，以及

Yes, I'll try to pick out sort of five of the top players right now when it comes to these models specifically.

是的，我會試著挑選出目前在這些模型方面，大概五個頂尖的廠商。

So first, Baidu, the biggest search engine in China, one of the biggest Internet companies.

首先是中國最大的搜尋引擎、也是最大的網路公司之一的百度。

It's got a foundational model called Ernie.

它有一個叫做 Ernie 的基礎模型。

No, it's like a mini game inside beyond the Valley this week.

不，這就像本週在「Beyond the Valley」裡的一個小遊戲。

So it's enhanced representation through knowledge integration.

所以它是透過知識整合來增強表徵。

所以 Ernie 的結尾沒有 E，是嗎？

對，沒有，他們就是這樣取名字的。

But it is Ernie was among among the first major ones to come out in China.

但 Ernie 是中國最早出現的主要模型之一。

There's been multiple versions as they've upgraded it.

隨著他們升級，已經推出了多個版本。

it has capabilities on par with GPT4, which is one of the latest models from OpenAI that's according to Baidu, of course, that's the claims they're making.

它擁有與 GPT4 相媲美的能力，GPT4 是 OpenAI 最新的模型之一，當然，這是百度方面的說法。

And this Ernie foundational model underpins Ernie Bot, which is Baidu's chatbot.

而這個 Ernie 基礎模型是 Ernie Bot 的基礎，Ernie Bot 是百度的哈拉機器人。

So similar to ChatGPT, ask it a question, it gives you an answer, but

所以，就像 ChatGPT 一樣，你問它一個問題，它會給你一個答案，但是

it's in Chinese and so it's very key for the China market.

它是中文的，所以對中國市場非常重要。

Second big player, Alibaba, the E commerce and cloud giant which we've spoken about before, its foundational model called Tongi Tsianyuan, which is shortened obviously occasionally to Qwen Alibaba has developed various versions of this, which is interesting and they can carry out different tasks rather than this sort of all encompassing model.

第二個大咖，阿里巴巴，我們之前談過的電子商務和雲端巨頭，它的基礎模型叫做「通義千問」，簡稱 Qwen。阿里巴巴開發了這個模型的各種版本，這很有趣，而且它們可以執行不同的任務，而不是這種包羅萬象的模型。

It's got one model for example, that's focused on creating content or solving math problems and another one that does something else.

例如，它有一個模型專注於內容創作或解決數學問題，另一個模型做別的事情。

So it's gone for a more specific approach here.

所以它在這裡採取了更具體的方法。

Another one for example can understand audio based inputs and give text based outputs.

例如，另一個模型可以理解基於音訊的輸入並給出基於文字的輸出。

So various different models that they have Tencent, the third biggest player, they have a foundational model called Hunyuan.

所以它們有各種不同的模型，第三大咖騰訊，它們有一個基礎模型叫做「混元」。

And Tencent has said that this model has strong Chinese language processing abilities and advanced logical reasoning.

而騰訊表示，這個模型擁有強大的中文處理能力和先進的邏輯推理。

This reasoning part is something we hear a lot about disability for AI models to actually understand and then reason and give that response.

這個推理能力是我們常聽到的，AI 模型在理解後進行推理並給出回應的難點。

As a result, the model can support functions that include things like image generation, text recognition.

是以，該模型可以支援包括圖像生成、文字識別等功能。

Again, this is something we've seen across the models and Tencent has actually we've spoken, I think we've probably touched upon Tencenter episodes.

這點我們在其他模型上也見過了，騰訊其實，我認為我們可能已經談過騰訊的相關內容了。

It runs the biggest messaging service in China, WeChat.

它營運著中國最大的通訊服務 WeChat。

It's also one of the biggest gaming companies in the world and it also is starting to dip its toe into E commerce.

它也是全球最大的遊戲公司之一，並且也開始涉足電子商務。

So it's positioned this model is one that can be used by companies in those industries in gaming and E commerce.

所以它將這個模型定位為遊戲和電子商務等行業的公司可以使用。

So everyone's trying to differentiate here.

所以大家都在這裡試圖做出差異化。

It's also got its own chatbot, Yuan Bao and this again very different, can pull information from WeChat, which has lots of information on it as Tencent, like I said, looks to differentiate from rivals.

它還有自己的哈拉機器人 Yuan Bao，這同樣非常不同，可以從 WeChat 獲取資訊，而 WeChat 上有很多資訊，正如我所說，騰訊正試圖與競爭對手區分開來。

Two more Huawei, huge tech giant which we've spoken a lot about.

還有另外兩家，華為，我們談論了很多的科技巨頭。

Its foundational model, Pangu Huawei has taken again a slightly different approach and it's made various different AI models specifically tailored to industries.

華為的 Pangu 模型採取了稍微不同的方法，它製作了各種專為產業量身打造的 AI 模型。

So it's got an AI model for finance, one for manufacturing, one for the government, one for mining, one even for meteorology.

所以它有金融業的 AI 模型、製造業的、政府的、礦業的，甚至還有氣象的。

And so it's trying to get very specific with the models for what it sees as some of the key industries.

是以，它試圖針對它認為的關鍵產業，讓模型變得非常具體。

And those are sold through its cloud business, for example.

這些模型是透過它的雲端業務銷售的，例如。

And it can also support features, it says, such as the ability to generate code and virtual human avatars, which is actually a super interesting space in China.

它還能支援生成程式碼和虛擬人類化身等功能，這在中國是一個非常有趣的領域。

The ability to just generate a kind of character that can speak and reason and understand things.

能夠生成一個可以說話、推理和理解事物的角色。

Final company, ByteDance, the owner of TikTok, it's got a foundational model called Doubao and it came to market not that long ago with a super cheap price point and kind of caused a bit of a price war in China.

最後一家公司是 TikTok 的母公司 ByteDance，它有一個名為 Doubao 的基礎模型，不久前才上市，價格非常便宜，並在中國引發了一場價格戰。

So those are, I'd say some of the biggest is a bunch of other startups in China doing very interesting things with foundational models.

所以這些，我認為是一些最大的公司，中國還有許多其他新創公司在基礎模型方面做著非常有趣的事情。

But those are some of the big names right now.

字幕列表影片播放

像英偉達（Nvidia）這樣的美國公司備受關注。但中國在人工智能領域做了些什麼？ (U.S. firms like Nvidia get the attention. But what's China doing in AI?)

specific

sort

process

access