跟讀練習: We Can Now Simulate a Human Brain, Scientists Show - 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。

85 句

如果句子過短或過長，請點選 Edit 進行調整。

Supercomputers are now good enough to simulate a full human brain.

That sounds like science fiction, but that's what scientists showed in a recent paper.

Haven't we tried this before and it didn't work?

Kind of, but this time it's different.

Computers have much advanced and the scientists have come up with better ways to calculate what human brains do.

How?

And even if we could simulate a human brain, should we?

Let's have a look.

The current models for artificial intelligence use computer code to simulate neural networks,

which are motivated by the functions of neurons in real brains.

But these AIs are far from the real thing.

The biggest difference may be that real brains are extremely differentiated in function.

They have areas for different tasks.

Scientists have tried to reproduce the function of real neural nets on a computer for decades.

They started with a small worm whose brain has just about 300 neurons.

The first simulations could reproduce simple movements of the worm.

By 2023, science and technology had advanced enough to fully simulate the brain of a fruit fly with about 140,000 neurons.

The Allen Institute for Brain Science, meanwhile, has been working on putting a mouse brain on a supercomputer.

The entire mouse brain has something like 70 million neurons.

They are now at a few million, and that runs on a supercomputer in Japan.

The new paper now outlines a new method that'd be a big and sudden jump forward, almost reaching the capacity to simulate the human brain.

The human brain has about 80 billion neurons.

In the paper, they write that with their new method,

network sizes of 2 times 10 to the 10 neurons can be reached with our approach on the upcoming exascale supercomputer Jupiter.

So that'd be 20 billion neurons with the next upcoming supercomputer.

It's fair to expect then that the next generation of supercomputers will be able to simulate the entire human brain.

What is this new method?

Basically, they have removed a bottleneck in the way that big neural networks are currently simulated.

With the previous method, one would instantiate the entire network on the entire computing cluster.

With the new method, they say we give some 100,000 neurons to each GPU, let each GPU deal with those, and then connect the GPUs.

That is, this is a massively parallel and local approach that avoids a lot of shoveling around of information, so it's computationally much leaner.

The maths is then pretty straightforward.

In detail, they estimate that a single NVIDIA A100 GPU could deal with about 225,000 neurons.

The Leonardo supercomputer in Italy has about 14,000 GPUs, so that'd make about a few billion neurons.

Jupiter is a new supercomputer cluster currently under development in Eulich in Germany.

Each GPU could manage 800,000 or so neurons and a total of 20 billion.

That seems big, but wait, wasn't there a similar project a decade ago?

Yes, there was the Human Brain Project, an international effort but centered in the EU.

It ran from 2013 to 2023, had over a billion euros in funding and aimed to build a digital human brain too.

They wanted to simulate a few million neurons, but it was a disaster.

The project leadership had this great ambition of simulating parts of the brain, but it was somewhat unclear what this would be good for.

Neuroscientists wanted to study more specific details of processes in certain parts of the brain.

That led to a quite big rift in the community and in the end, to make a long story short, Not much came out of it, besides a lot of reports.

Some of the authors of the new paper also worked on software for the previous human brain project.

So what's different this time?

In a nutshell, more neurons, fewer neuroscientists.

I think these people just want to simulate a human brain to see what it does.

They don't have any particular plan for what to do with it.

And you know, I can totally see the point of doing this.

What could such a simulation do for us?

I don't know, but we can talk about some things it wouldn't do.

We actually don't have the exact map of any human neural network that's called the connectome.

So we can't build an entire human brain because we don't know exactly what it looks like.

Without doing that first, a computer simulation could at best be a guess for how the brain is connected,

and it's somewhat unclear how much it could do.

Another issue is that it'd be difficult to train such an artificial human brain

because we'd first have to figure out how to convert input into neural signals.

We know some of that, but only in rough terms.

And then there is the moral question.

Should we simulate a human brain?

What if it does actually think?

What if it suffers?

What if the EU Commission says we're not allowed to turn it off and it blocks the only German supercomputer cluster?

Personally, I think it'd well be worth it.

It'd demonstrate that there's at least one functioning brain in Europe.

I read a lot of news to keep you well informed

and I found that the best way to do this is on Ground News, who've been sponsoring this video.

Ground News is a news platform that collects and summarizes news which has been published all over the world.

Not only do they collect all articles on the same story in one place and give you a quick summary,

they also give you a lot of extra information that you don't find in the standard media.

Take for example this recent story about scientists studying polarize.

You can see right away that this basically wasn't covered on the political right as almost all climate change stories.

Ground News gives you a quick overview on the most important points.

And here you can see who owns the media outlets.

You get a factuality check for each news item and you can see where the story has appeared.

Ground News also has this great feature called Blindspot.

This tells you which news has been almost exclusively covered only by one side of the political spectrum.

And of course, I have a special offer for you.

That's a 40% discount on the Vantage plan, which gives you access to all their features.

All you need to do is use my link ground.news.sabina or use the QR code so they'll know I sent you.

I really think that Ground News is onto something with their news overview.

So go and have a look.

Thanks for watching.

See you tomorrow.

Thank you.

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

這段影片共有 85 個句子、1106 個單詞可供跟讀。講話部分時長為 6:59。說話人語速自然，每分鐘約 158 個詞，接近日常對話。 82% 的單詞屬於英語最常用的 3,000 詞，其餘的詞建議在練習前先學一下。

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
scientist 名詞	/ˈsaɪəntəst/	科學家 /科學家
cluster 名詞	/ˈklʌstɚ/	串, 叢 /叢
simulation 名詞	/ˌsɪm.jəˈleɪ.ʃn̩/	模擬 /模擬
connect 動詞	/kəˈnɛkt/	連線 /連線
collect 動詞	/ˈkɑlɪkt/	庋藏, 收藏
mouse 名詞	/maʊs/	лочў, 鳥鼠 /鳥鼠
artificial 形容詞	/ˌɑː.tɪˈfɪʃ.əl/	人造的
overview 名詞	/ˈoʊ.vɚˌvjuː/	概觀 /概觀
unclear 形容詞	/ʌnˈklɪɚ/	不清
worm 名詞	/wɝm/	蠕蟲 /蠕蟲
reproduce 動詞	/ˌɹi.pɹəˈdus/	複製 /複製
meanwhile 副詞	/ˈmiːnwaɪl/	同時 /同時
moral 形容詞	/ˈmoɹəl/	道德的
suffer 動詞	/ˈsʌfɚ/	受苦, 遭受
fiction 名詞	/ˈfɪkʃən/	虛構文學 /虛構文學, 小說 /小說

影片中出現的短語動詞

單詞	釋義
come up with 動詞	想出, 提出
figure out 動詞	弄清楚

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
現在完成進行時 have/has been + -ing — 以前開始、現在仍在繼續的動作	has been working · who've been sponsoring
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 we tried · have come · have tried
被動語態 be + 過去分詞 — 強調發生了什麼，而不是誰做的	can be reached · is connected · has been published

需要注意的發音

說話人用了 21 次縮略和弱讀形式，例如 don't, it'd, that'd。請按聽到的簡短形式來說。

“sh” 和 “zh” 音: simulation /ˌsɪm.jəˈleɪ.ʃn̩/, artificial /ˌɑː.tɪˈfɪʃ.əl/, fiction /ˈfɪkʃən/, ambition /æmˈbɪʃ.ən/, differentiate /dɪf.əˈɹɛn.ʃi.eɪt/
長單詞——注意重音位置: artificial /ˌɑː.tɪˈfɪʃ.əl/, exclusively /ɪkˈskluːsɪvli/, mathematics /mæθ(.ə)ˈmæt.ɪks/, differentiate /dɪf.əˈɹɛn.ʃi.eɪt/, computationally /ˌkɑmpjʊˈteɪʃənəli/

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 scientist, cluster, simulation 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: We Can Now Simulate a Human Brain, Scientists Show - 通過影片學習英語口語

選擇練習模式

下載應用

同頻道

Make AI Believe in God to Keep Humans Safe, Scientist Argues

The Uncertainty Principle: What Does It Mean, How Does It Work?

Mathematician Collapses All Functions to One Weird Formula

AI Will End Philosophy, Physicist Claims

Has quantum mechanics proved that reality does not exist?

Does God exist? Science does not have an answer.

Superfluid Dark Matter

What is time?

The End of Human Mathematics? AI's Recent Breakthrough -- And What's Next

FINALLY: A Proof that Particles Take All Paths At Once?

10 things you should know about black holes

Our Solar System Lost Two Planets!

I was SO wrong about quantum computing.

推薦影片

The Engineering that Runs the Digital World 🛠️⚙️💻 How do CPUs Work?

Your Brain Peaked at Birth

What AI can (and can't) currently do in a real lab | Jennifer Doudna

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

Would You Upload Your Mind?

The Insane Biology of: The Octopus

The Limitless AI Lie. The Bubble Is Slowly BURSTING.

The different levels of how Claude thinks

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

How Social Media Is Rewiring Our Brains