跟讀練習: Did the AI bubble just pop? - 通過影片學習英語口語

選擇練習模式

每種模式都會開啟全屏練習室。

71 句

如果句子過短或過長，請點選 Edit 進行調整。

just three weeks ago, it felt like nothing could go wrong.

SpaceX was up 7 % in a day,

every tech CEO in America was on stage explaining why they needed another $100 billion to decimate farmland in rural America,

and Jensen Wong was telling anyone who'd listen that demand for AI chips was insatiable.

But since then, the market took a dump.

Meta created an entirely new class of non -voting stock just to pay for GPUs,

while Microsoft offered a quarter of its employees money to leave so it can afford more of them.

Elon predicted that 200 NeoClouds are about to experience Guppy -level death, and CoreWeave, a former Ethereum mining operation

that became the poster boy of reckless AI spending after they pivoted into renting NVIDIA GPUs back to NVIDIA,

is now worth a third of what it was at the beginning of the month.

And then on Monday, Anthropic filed to go public, which showed that between January and when the government banned Fable back in June,

they lost more money than the annual GDP of over half the countries on Earth.

So naturally, the question on everyone's mind is, are we finally cooked?

In today's video, we'll break down the roughest three weeks in the history of AI,

find out what a neocloud is and why Elon wants them dead, and we'll analyze Anthropics financials in their Trust Me Bro S1.

It is October 28th, 2026, and you're watching the Code Report.

Before we can talk about the AI bubble, you need to understand how a GPU actually gets paid for.

And to do that, here's me in a bubble bath to explain.

First, NVIDIA makes a GPU that they need to sell.

The problem is the GPU is expensive and you don't have enough money to buy it.

So NVIDIA says, hey, we have money

and we'll give it to you as long as you give us a portion of your company

and use the money we give you to buy our GPUs.

But NVIDIA's money only covers a fraction of what you need.

So you go to Wall Street and borrow the rest by putting up the GPU you haven't bought yet as collateral.

Now normally, a bank wouldn't lend to you with a depreciating pile of chips as collateral,

but NVIDIA has promised that if nobody rents the GPUs they give you, they'll rent them themselves.

So NVIDIA's money gets used to buy NVIDIA GPUs that eventually get rented back to NVIDIA, which makes NVIDIA's stock go up,

which it uses to invest in more companies, which borrow more money to buy more NVIDIA GPUs, which keeps the good times going.

On Wall Street, they called this a circular financing arrangement, and back in college, we'd call it a s*.

The most popular company eating this forbidden fruit is CoreWeave, but NVIDIA has done it with a few other companies as well.

And to be honest, it works pretty well as long as the chips rent for more than they depreciate.

The labs keep buying, and someone keeps lending you money to buy more chips.

But three weeks ago, the third one stopped being true when it was discovered that CoreWeave, after it borrowed $52 billion to buy chips,

was now paying 15 % interest on that debt.

Then on Sunday night, Elon chimed in, saying that debt markets have almost entirely shut off for any company whose only revenue is AI data centers,

and that he expects 200 plus of these neocloud type companies to die.

But conveniently, SpaceX is fine, because it also happens to sell rockets.

And right in the middle of all this, Anthropic decided to ask the public for $100 billion at a valuation

that would make it worth more than 492 out of the 500 companies in the S &P 500,

with 4 out of those 7 remaining companies being where they rent their compute from

and happen to owe $265 billion to through 2045.

The S1 itself is 330 pages long, with 90 of those being risk factors to the company, which is the fun part.

In it, Anthropic admits that everything we've already covered over the last few months is true, including Mythos being allergic to its sandbox,

Claude Haiku filing a fake homicide tip through the Philadelphia Police Department's website,

and a Chinese state -sponsored group using Claude code to attack banks and government agencies.

The filing also says that Claude now uses itself to improve itself,

and that they might not notice their models getting significantly smarter until after it's already happened.

Their guiding principle, according to the foreword, is to under -promise and over -deliver,

which is a little worrisome for a company whose head of alignment was quoted just last month,

saying he thinks there's a greater than 10 % chance that AI kills all humans within the next decade.

So are we cooked?

Well, as always, it depends.

It's not looking great

if your whole business model is borrowing money from one company just to give it right back to them.

But it's hard to call AI a bubble the day after

OpenAI claims one of its models disproved a 27 -year -old conjecture that half of complexity theory is built on.

But thankfully, as developers, things are actually looking pretty decent right now, since every one of these companies are fighting to sell us subsidized tokens.

Akimi just dropped an open model within a point of Opus at half the price that we talked about yesterday,

and the GPU score we financed should be on eBay any week now.

And if you'd like to be one of the people profiting from the exit liquidity instead of providing it,

you can put those subsidized tokens to work with HyperAgent, the sponsor of today's video.

If you're already paying for ChatGPT, you can now use that plan to sign into HyperAgent

and hire a team of agents without setting up yet another AI subscription.

Right now, I'm using it to work on a v4 release of a date library I'm contributing to.

So I set up a HyperAgent room, which is basically a Slack channel where all my agents and human co -maintainers can work together.

I've got one agent triaging new issues as they come in another one drafting pull requests for new features, and a third one writing the changelog.

They all work in parallel and hand things off to each other, and only ping a human when something needs approval.

And anyone on a team can jump in to ask questions or make changes whenever they want, instead of routing everything through me.

This has been The Code Report, thanks for watching, and I will see you in the next one.

下載應用

流利口語所需的一切

AI 發音評分逐句打分

音標練習掌握每個音

詞彙學習建立你的詞庫

詞彙遊戲邊玩邊學

掃碼

Download on theApp Store Get it onGoogle Play

同頻道

FOR YOU

本課的詞彙與口語要點

這節口語課以影片“Did the AI bubble just pop?”為素材。影片中反覆出現的詞有：Gpu, Nvidia, companies, chips。這段影片共有 71 個句子、1128 個單詞可供跟讀。講話部分時長為 5:22。說話人語速較快，每分鐘約 210 個詞，會出現較多連讀和弱讀。只有 81% 的單詞屬於英語最常用的 3,000 詞，詞彙難度較高。

影片中的重點詞彙

影片中 15 個值得學習的單詞，附發音和釋義：

單詞	發音	釋義
chip 名詞	/t͡ʃɪp/	籌碼
bubble 名詞	/ˈbʌbəl/	泡, 氣泡 /氣泡
borrow 動詞	/ˈbɑɹ.oʊ/	借, 貸 /貸
sponsor 名詞	/ˈspɒn.səː/	保證人 /保證人
lend 動詞	/ˈlɛnd/	借出, 借
token 名詞	/ˈtoʊkən/	代幣 /代幣, 籌碼 /籌碼
collateral 名詞	/kəˈlætəɹəl/	擔保 /擔保, 抵押品
bath 名詞	/bɑːθ/	浴缸, 浴盆
decent 形容詞	/ˈdiː.sənt/	體面 /體面, 正派
hire 動詞	/ˈhaɪɹ/	租借, 出租
discover 動詞	/dɪˈskʌvə/	發現 /發現, 發覺 /發覺
exit 名詞	/ˈɛɡzət/	離開 /離開, 退場 /退場
contribute 動詞	/kənˈtɹɪb.jut/	貢獻 /貢獻
principle 名詞	/ˈpɹɪn.sɪ.pəl/	原理, 原則 /原則
invest 動詞	/ɪnˈvɛst/	耗費 /耗費, 投資 /投資

影片中出現的短語動詞

單詞	釋義
break down 動詞	壞了 /壞了, 出故障
find out 動詞	找出, 查明
go up 動詞	上昇 /上升, 上去

影片中的語法

說話人最常用的結構，並附上影片中的原話：

結構	影片中的用法
現在完成時 have/has + 過去分詞 — 過去發生但與現在仍有關聯的事	haven't bought · has promised · has done
被動語態 be + 過去分詞 — 強調發生了什麼，而不是誰做的	was discovered · was quoted · is built
定語從句 who / which + 從句 — 補充說明人或事物	NVIDIA, which makes · company, which is · room, which is

需要注意的發音

說話人用了 16 次縮略和弱讀形式，例如 we'll, I'm, you're。請按聽到的簡短形式來說。

“sh” 和 “zh” 音: subscription /səbˈskɹɪpʃən/, fraction /ˈfɹæk.ʃən/, valuation /ˌvæ.ljuːˈeɪ.ʃən/, depreciate /dəˈpriʃiˌeɪt/, insatiable /ɪnˈseɪʃ(j)əbəl/
長單詞——注意重音位置: collateral /kəˈlætəɹəl/, naturally /ˈnæt͡ʃ.(ə.)ɹə.li/, developer /dɪˈvɛləpɚ/, complexity /kəmˈplɛk.sɪ.ti/, valuation /ˌvæ.ljuːˈeɪ.ʃən/

中文母語者容易讀錯的音：

/v/ — 不要讀成 /w/，上齒輕觸下唇: discover /dɪˈskʌvə/, invest /ɪnˈvɛst/, developer /dɪˈvɛləpɚ/, valuation /ˌvæ.ljuːˈeɪ.ʃən/, conveniently /kənˈvinjəntli/
詞尾子音 — 要讀清楚，後面不要加母音: chip /t͡ʃɪp/, lend /ˈlɛnd/, bath /bɑːθ/, lab /læːb/, decent /ˈdiː.sənt/
/r/ 和 /l/ 的區別: collateral /kəˈlætəɹəl/, naturally /ˈnæt͡ʃ.(ə.)ɹə.li/, principle /ˈpɹɪn.sɪ.pəl/, parallel /ˈpæ.ɹəˌlɛl/, circular /ˈsɜɹ.kjə.ləɹ/

如何用這段影片練習

先完整聽一遍影片，不要開口，記下不認識的單詞。
先用 0.75 倍速逐句跟讀，熟練之後再回到正常速度。
錄下自己的聲音並與原聲對比，特別注意 chip, bubble, borrow 這類單詞。

什麼是跟讀法？

跟讀法 (Shadowing) 是一種有科學依據的語言學習技巧，最初開發用於專業口譯員的培訓，並由多語言者Alexander Arguelles博士普及。這個方法簡單而強大：您在聽英語母語原聲的同時立即大聲重複——就像是一個延遲1-2秒緊跟說話者的影子。與被動聽力或語法練習不同，跟讀法強迫您的大腦和口腔肌肉同時處理並模模擬實的講話模式。研究表明它能顯著提高發音準確性，語調，節奏，連讀，聽力理解和口語流利度——使其成為雅思口語備考和真實英語交流最有效的方法之一。

影子跟讀法: 閱讀完整分步指南 →

跟讀練習: Did the AI bubble just pop? - 通過影片學習英語口語

選擇練習模式

下載應用

Brainf**k in 100 Seconds

PostgreSQL in 100 Seconds

React Native in 100 Seconds

Did Google just dethrone OpenAI and Anthropic?

A $6.3 billion open-weight model just got embarrassed by the French...

Every app on Earth is now open source... whether it likes it or not

OpenAI’s lead engineer was secretly a spy...

The most interesting "hack" in history...

Mind-bending new programming language for GPUs just dropped...

China just dropped a new frontier model... and nobody can explain how

The last real programmers are dying out...

Intel chips can’t possibly be this bad… 100% crash rate?

Docker in 100 Seconds

OpenAI’s new browser feels familiar…

Firefox and Tor hit with 9.8 critical level exploit

5 open source tools that replaced my $320/mo AI stack...

C++ in 100 Seconds

Rust in 100 Seconds

Linux in 100 Seconds

The plan to break apart Google... RIP Chrome

No, Really — Is A.I. a Bubble? | The Ezra Klein Show

Expensive RTX 5090 for LLMs? NO. Use This Instead. (Tesla V100 SXM2 + Z8 G4, #RACERRRZ)

Why AI Tokens are so Expensive - Computerphile

實戰練英文｜黃仁勳：你們要加油，不然我會拿走所有機會，然後我會變很有錢｜德國沒有下一個機會了｜投資德國高峰會｜長文、註解、大字版、中英對照

AI Governance Explained: Ethical, Transparent & Accountable AI for the Future

How AI is helping brands identify consumer trends faster

Jack Dorsey’s New Dessa AI Voice Sounds Scary Human

The Limitless AI Lie. The Bubble Is Slowly BURSTING.

Elon Musk: The End of Work, the Death of Money, and a World Run by AI

The Only SaaS Moats Left in 2026