跟讀練習: How does AI actually work? - 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。

26 句

如果句子過短或過長，請點選 Edit 進行調整。

I'm confident that this may be the most important video you'll see in a while.

Let's talk about how AI actually works because it will shape how you use it, how you think about it, and make informed decisions about it.

I'm a Stanford -trained computer scientist for AI who's been building AI systems for over a decade.

I'm teaming up with Anthropic, the team behind Claude, for a series, and today we're looking under the hood.

AI agents are made up of two parts, the large language model, LLM, and the hardest this.

In this video, we discussed the word generating LLMs, and today we'll discuss the agent harness.

So we have the word generator, the core brain, i .e whatever version of sonnet, opus, fable you select when using Claude.

And then we have the harness, which is all of the code that enables that word generator to take actions.

So remember from the last video that everything the model knows about you is just added info into its prompt.

The harness is what manages

that memory your history your files your context it also gives the model access to external tools like gmail

or canva and handles sub -agent orchestration

which is delegating tasks to other agent instances sometimes in parallel

and here's how the two pieces work together modern llms are trained to reason to think through steps

and decide on how to act when the model wants to use a tool, it outputs a tool call, essentially structured text saying run this,

and then the harness executes it.

Then the output of the actions gets placed into the model's prompt behind the scenes, and the model then reflects on how the tool call went, and the loop continues.

It thinks again on what to do, maybe calls a tool, and it goes on and on.

Note that LLMs have a limit to their working memory called the context window.

Your chat history, your prompt, files, all those tool outputs all go in to that context window

and eventually it fills up and the harness detects this and intelligently performs context compression.

So those are the two pieces of chatbot AI agents.

LLMs are the brain generating text and the harness is like the body.

It gives them the ability to actually take actions and make changes.

I hope this is helpful.

Please leave your questions in the comments and follow for more AI explainers.

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

FOR YOU

本課的詞彙與口語要點

這段影片共有 26 個句子、409 個單詞可供跟讀。講話部分時長為 2:17。說話人語速自然，每分鐘約 179 個詞，接近日常對話。只有 80% 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
output 名詞	/ˈaʊtpʊt/	輸出 /輸出
generator 名詞	/ˈdʒɛnəɹeɪtə(ɹ)/	發電機 /發電機
select 動詞	/sɪˈlɛkt/	選 /選, 挑選 /挑選
helpful 形容詞	/ˈhɛlp.fl̩/	有幫助 /有幫助
essentially 副詞	/ɪˈsɛnʃəli/	本質上 /本質上, 基本上
reflect 動詞	/ɹɪˈflɛkt/	反映, 反射
loop 名詞	/luːp/	圈, 環 /環
parallel 形容詞	/ˈpæ.ɹəˌlɛl/	平行的
scientist 名詞	/ˈsaɪəntəst/	科學家 /科學家
inform 動詞	/ɪnˈfɔɹm/	通知, 告知
detect 動詞	/dɪˈtɛkt/	察覺 /察覺, 發覺 /發覺
execute 動詞	/ˈɛksɪˌkjuːt/	處決 /處決, 處死 /處死
delegate 名詞	/ˈdɛlɪɡət/	代表, 使節 /使節
chatbot 名詞	/ˈtʃætbɑt/	聊天機器人 /聊天機器人
fable 名詞	/ˈfeɪbəl/	寓言

影片中出現的短語動詞

單詞	釋義
fill up 動詞	裝滿 /裝滿

需要注意的發音

說話人用了 6 次縮略和弱讀形式，例如 I'm, we'll, we're。請按聽到的簡短形式來說。

“sh” 和 “zh” 音: essentially /ɪˈsɛnʃəli/, compression /kəmˈpɹɛʃ.ən/
長單詞——注意重音位置: generator /ˈdʒɛnəɹeɪtə(ɹ)/, essentially /ɪˈsɛnʃəli/

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 output, generator, select 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: How does AI actually work? - 通過影片學習英語口語

選擇練習模式

下載應用

推薦影片

When AIs act emotional

LangChain in 5 Minutes (Explained Clearly)

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

AI, Machine Learning, Deep Learning and Generative AI Explained

Agentic AI Frameworks Explained: Workflows, Multi-Agent, & Production

Jack Dorsey’s New Dessa AI Voice Sounds Scary Human

What is a PLC and how does it work?

Dear Higgsfield, I reverse engineered you in 12 hours

Why AI Tokens are so Expensive - Computerphile

AI Will End Philosophy, Physicist Claims

AI harness crash course

Do Elite Degrees Still Matter In The AI Era? | Jamie Beaton from Crimson Education | EP 410