跟读练习: Did Anthropic Kill Git? - 通过视频学习英语口语

选择练习模式

每种模式都会打开全屏练习室。

65 句

如果句子过短或过长，请点击 Edit 进行调整。

Did Anthropic kill Git?

This week, a lot of people on X decided it did.

One post with over 600 ,000 views says Anthropic has replaced Git internally and the titans are falling.

Another says this is much bigger than it looks.

It all traces back to one post from Anthropic's own account, seen more than 6 .5 million times.

Part of the story is true.

Anthropic did stop using Git, but only for its own code, inside its own company.

Git itself is fine.

Anthropic keeps almost all of its code in one giant repository, a monorepo.

In May, about 2 ,500 changes were merged into it every day.

By September, it was over 30 ,000 a day.

On September 9th alone, 30 ,401 changes went in.

Their chart even has a dip in late July labeled the quiet week, and right after it, the line goes almost straight up.

Claude writes a lot of that code, and Anthropic says every bottleneck they remove leads to even more of it.

At that rate, the repository got heavy.

The history of the main branch alone is almost 100 gigabytes.

Their CI system checks out the code about 25 million times a day before you count developers, agent sandboxes, and research jobs.

Anthropic did the maths.

If every one of those checkouts fetched the repo the normal way, that would be over 600 petabytes of Git traffic every day.

Git was designed for a different kind of user.

Every clone is a full copy of the repository with all of its history, and it can do almost everything offline.

That's great for a person on a laptop on a plane.

At Anthropic, almost every client is an agent or a CI job, and those are always online.

They tried the usual fixes first, like shallow clones and sparse checkouts.

At the peak, they were running several hundred internal Git mirrors just to keep up.

So they built their own.

It's called Scion, and it moves the work from your machine to the server.

Commits, merges, rebases, blame, and history all happen on the server so there's nothing to push.

The client keeps a small cache and only downloads the files it actually reads.

A CI job that needs a diff can ask the server for it and never check out the code at all.

Between the client and the server sit two layers of caches, one on every machine and one in every region.

And each layer answers over 99 % of requests by itself.

Anthropic says it borrowed ideas from Google's Piper and Meta's Sapling, which were built for huge human teams.

On a developer machine, Git status took almost four seconds, and Scion does it in 19 milliseconds.

For agents, setting up a fresh work tree went from seven minutes to five seconds.

The whole system uses about 200 times less CPU and memory than the old mirrors, and CI got a quarter of its CPU back.

These are Anthropic's own measurements, and nobody outside the company has checked them yet seven people built it.

In five months, they made Scion, a code review tool called Antique, and a few other systems,

and they moved the whole monorepo over while the merge rate went up 10 times.

Claude wrote a lot of that code too.

The team had handwritten a pipeline to import the old Git history,

and an early version of Claude Mythos rewrote it to use 20 times less memory and run four times faster.

Every engineer on the project now rates Claude as better at writing code than they are.

They also say a big part of their job became telling Claude no

when it took shortcuts or piled on checks nobody needed.

On Hacker News, the thread passed 800 points and 900 comments in a day.

The most replied comment says that every time the writer stops reviewing Claude's code, they later find janky crap.

Others said a central server you always have to reach sounds a lot like subversion or perforce.

The older systems get replaced.

One person asked why Anthropic didn't just split up the monorepo.

On X, Dex Horthy pointed at the trade -off most people will feel.

Today, if GitHub is down, you can still commit on your laptop.

With Scion, if the server is down, you can't.

So did Anthropic kill Git?

No. In their own words, Scion is by no means a suitable replacement for Git for general use.

They say it only makes sense for truly enormous repos, with tens of thousands of engineers or swarms of agents.

It isn't open source and they don't plan to release it.

Your repos on GitHub don't change, and Claude code still works with plain Git.

A small team has already posted an unofficial open source version, but Anthropic has nothing to do with it.

What Anthropic wants is for Git to pick up some of these ideas,

like servers that do more of the work and answer questions about history through an API.

They say they want to build that with GitHub, GitLab, and the open source community,

and they expect other companies using AI coding tools to hit the same wall within a year.

If that happens, the changes will most likely land on the server side,

and the Git on your laptop will keep working the way it does today.

下载应用

流利口语所需的一切

AI 发音评分逐句打分

音标练习掌握每个音

词汇学习建立你的词库

词汇游戏边玩边学

扫码

Download on theApp Store Get it onGoogle Play

FOR YOU

本课的词汇与口语要点

这节口语课以视频“Did Anthropic Kill Git?”为素材。视频中反复出现的词有：Git, Anthropic, code, server, Claude。这段视频共有 65 个句子、881 个单词可供跟读。讲话部分时长为 4:49。说话人语速较快，每分钟约 183 个词，会出现较多连读和弱读。只有 79% 的单词属于英语最常用的 3,000 词，词汇难度较高。

视频中的重点词汇

视频中 15 个值得学习的单词，附发音和释义：

单词	发音	释义
server 名词	/ˈsɝvɚ/	服務器 /服务器, 伺服器
laptop 名词	/ˈlæp.tɑp/	手提電腦 /手提电脑, 手提計算機 /手提计算机
merge 动词	/mɝd͡ʒ/	合并
mirror 名词	/ˈmiɹɚ/	җинзы, 鏡 /镜
layer 名词	/ˈleɪ̯ɚ/	層 /层
clone 名词	/kloʊn/	克隆, 複製 /复制
checkout 名词	/ˈt͡ʃɛkˌaʊt/	收款處 /收款处
cache 名词	/kæʃ/	緩存 /缓存, 快取
chart 名词	/t͡ʃɑɹt/	圖表 /图表
thread 名词	/ˈθɹɛd/	線 /线
label 名词	/ˈleɪ.bəl/	標籤 /标签, 簽條 /签条
plain 名词	/pleɪn/	平原, 平野
suitable 形容词	[ˈsjuː.tə.bɫ̩]	合適 /合适, 適當 /适当
enormous 形容词	/ɪˈnɔː.məs/	巨大的
trace 名词	/tɹeɪs/	痕跡 /痕迹

视频中出现的短语动词

单词	释义
check out 动词	退房
go up 动词	上昇 /上升, 上去
pick up 动词	拾取, 撿 /捡

视频中的语法

说话人最常用的结构，并附上视频中的原话：

结构	视频中的用法
现在完成时 have/has + 过去分词 — 过去发生但与现在仍有关联的事	has replaced · has checked · has already posted
被动语态 be + 过去分词 — 强调发生了什么，而不是谁做的	were merged · was designed · were built

需要注意的发音

说话人用了 5 次缩略和弱读形式，例如 don't, can't, didn't。请按听到的简短形式来说。

“th” 音: thread /ˈθɹɛd/, mathematics /mæθ(.ə)ˈmæt.ɪks/
“sh” 和 “zh” 音: cache /kæʃ/, measurement /ˈmɛʒ.ə.mənt/, shallow /ˈʃæloʊ/, unofficial /ʌn.əˈfɪʃ.əl/, shortcut /ˈʃɔːtkʌt/
长单词——注意重音位置: developer /dɪˈvɛləpɚ/, mathematics /mæθ(.ə)ˈmæt.ɪks/, internally /ɪnˈtɜːnəli/, unofficial /ʌn.əˈfɪʃ.əl/, repository /ɹɪˈpɒz.ɪ.t(ə.)ɹɪ/

中文母语者容易读错的音：

/v/ — 不要读成 /w/，上齿轻触下唇: server /ˈsɝvɚ/, developer /dɪˈvɛləpɚ/, subversion /sʌ̈bˈvɝ.ʒən/
词尾辅音 — 要读清楚，后面不要加元音: laptop /ˈlæp.tɑp/, merge /mɝd͡ʒ/, commit /kəˈmɪt/, checkout /ˈt͡ʃɛkˌaʊt/, cache /kæʃ/
/æ/ — 嘴张大，不要读成 /e/: laptop /ˈlæp.tɑp/, cache /kæʃ/, mathematics /mæθ(.ə)ˈmæt.ɪks/, shallow /ˈʃæloʊ/, janky /ˈd͡ʒæŋ.kiː/

如何用这段视频练习

先完整听一遍视频，不要开口，记下不认识的单词。
先用 0.75 倍速逐句跟读，熟练之后再回到正常速度。
录下自己的声音并与原声对比，特别注意 server, laptop, merge 这类单词。

什么是跟读法？

跟读法 (Shadowing) 是一种有科学依据的语言学习技巧，最初开发用于专业口译员的培训，并由多语言者Alexander Arguelles博士普及。这个方法简单而强大：您在听英语母语原声的同时立即大声重复——就像是一个延迟1-2秒紧跟说话者的影子。与被动听力或语法练习不同，跟读法强迫您的大脑和口腔肌肉同时处理并模仿真实的讲话模式。研究表明它能显着提高发音准确性，语调，节奏，连读，听力理解和口语流利度——使其成为雅思口语备考和真实英语交流最有效的方法之一。

影子跟读法: 阅读完整分步指南 →

跟读练习: Did Anthropic Kill Git? - 通过视频学习英语口语

选择练习模式

下载应用

推荐视频

The easy way to keep your repos tidy.

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

Is React Native Dead? Shopify Is Leaving

The Philosophy of Rick Sanchez

How Did Ancient Humans Survive the Deadliest Predators on Earth?

Go’s Big Mistake

Creator of C# and TypeScript: "AI will NEVER Replace Coders, Here's Why" | Anders Hejlsberg

Linux got wrecked by backdoor attack

GitHub Copilot now controls your command line...

Linus Torvalds Explains What Makes Linux So Fast

STOP Running Code On Your Computer

We Let $200 GPT-Pro Think for 7 Days on Unsolved Math