跟讀練習: The making of Anthropic - 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。
200 句
如果句子過短或過長,請點選 Edit 進行調整。
1
So why did you all start Anthropic?
2
It still feels like then, in that we knew a secret the rest of the world didn't know.
3
I think if you truly see what this is, you can't ever walk away.
4
I showed up at Princeton as a PhD student who wanted to do theoretical physics.
5
I was a professor at Johns Hopkins before Anthropic.
6
We're all envious of the great discoveries of the past.
7
My PhD thesis is in theoretical physics.
8
Out of college, I went and worked in global development.
9
I was a journalist before I did any of this, and I studied data centers obsessively.
10
It's going to get weird.
11
I guess I thought of myself as a mathematician and programmer.
12
I got into AI back in 2016, and at the time, like, an AI couldn't talk.
13
After I graduated, I went to Stanford and did my postdoc in cancer biomarkers.
14
From a very early age, Dario and I were really curious about how to make the world better.
15
I was very interested in questions of how we can use computer science as a tool that is good for the world.
16
I was a little bit frustrated with the physics research I was doing, and AI seemed exciting and like a breath of fresh air.
17
2013 or 2014, I first started seeing the big results in AI,
18
I then wondered if advancing this technology would be the best way to advance biology.
19
We saw this connection between the level of impact we were trying to have in the world
20
and AI's potential to really achieve a lot of those goals.
21
Back in 2018, the future co -founders of Anthropic were doing a lot of sort of different and disparate things.
22
A group of us all at OpenAI, we are working on training models, we're working on policy, we're working on interpretability, we're working on safety.
23
I started working on these language models where you try to predict the next word.
24
A lot of folks thought at the time, like, oh, you're going to need to teach the network explicitly how to plan
25
and how to reason and how to use tools.
26
Dario noticed the trend, which he called the big blob of compute.
27
Just basically making a bigger machine, making a bigger computer, and jamming more data into it.
28
When you put in, you know, more compute and more data, you get increasing intelligence, and it's on this sort of predictable line.
29
We were just seeing, like, each time you scale it up, these new behaviors emerge, these new abilities emerge.
30
At which point you realize, wow, if this continues, we might be on the precipice of, like, a totally new reality,
31
a totally new form of technology.
32
Human -level intelligence, human -level AI is not, like, 100 years in the future, but it's actually within our lifetimes.
33
These kinds of very powerful models could cause enormous strides in health care, in robotics.
34
Computer programs, novel scientific ideas, mathematical theorems.
35
Maybe we could reduce cancer rates by 90%.
36
Maybe heart disease, obesity, all of the major infections will just be a thing of the past.
37
Any of the problems of the world that are bottlenecked by human ingenuity, suddenly you have much more ability to solve them.
38
Seeing the potential progress in AI was exciting.
39
At the same time, I think it was scary.
40
If you're building something that's more powerful than yourself or could grow to be more powerful than yourself, the potential risks of that are enormous,
41
no matter how you think about it.
42
I don't think I understood the full picture yet.
43
I don't think anyone did, but I definitely believed that there could be big safety and societal effects.
44
This was a technology that if it had any analogy at all, which I think it mostly didn't,
45
it would be more analogous to pharmaceuticals or unlocking the secrets of the atom.
46
Things that have immense value, but also can go wrong.
47
We really needed to build it in the right way.
48
Just this huge responsibility that you feel.
49
I'm one of the very few people in the world that understands what the consequences of this technology are going to be.
50
The thing we most wanted to do was work together in an environment with this deep commitment to safety.
51
We decided we had to build Anthropik
52
because we had to build an organization fit for the demands this technology was going to place on all of us.
53
One of the first big decisions we had to make was just what kind of company we wanted to be.
54
It was a more involved and complicated discussion than you might expect.
55
We knew that we had to be able to build some of the largest computers on the planet,
56
train frontier technology, find ways to deploy that technology into the world,
57
and be honest about the difficult aspects of the technology as well as its benefits.
58
And we eventually landed on Public Benefit Corporation.
59
That would set us up as a company that was not just fully governed by economic incentives,
60
but also by the incentives to do right by the world.
61
There are going to be times
62
when we're going to want to prioritize the mission at the cost of profits
63
or at the cost of providing the maximum value to a shareholder.
64
So we incorporated as a public benefit company on January 26, 2021, and then it was off to the races, time to build this thing.
65
This is a list of names people proposed for what we could call anthropic.
66
Here are some names I listed, apparently.
67
Squirrel, Owl, Whale, Differentiated, Humane.
68
Anthropic was contributed by Jared.
69
I realized that since Anthropic means kind of human -oriented or human -centered, that it seemed like a good name for an AI company
70
that wants AI to be developed in the best interests of humanity.
71
Then we looked at domain names that we could get.
72
Anthropic .com was, I think, by far the cheapest.
73
I thought that was a pretty good argument to go with Anthropik.
74
And there was a comment in the spreadsheet that says, we like the name, it is good.
75
And is Domain available?
76
Yes.
77
It was so chaotic at the beginning.
78
Everyone was doing extremely unglamorous schlep.
79
Tom was the one who had to go get us the AWS clusters.
80
Me and Chrisola built the first website.
81
Dario focused on fundraising and I was working on recruiting.
82
And there was briefly a few months where Jared Kaplan, and our chief scientist ran payroll.
83
That tells you the level of professionalism of the early operation.
84
I think there was a lot of work about sort of
85
just getting oriented to what kinds of things exist in a language model.
86
You find all these just like beautiful things inside that we didn't know about going in.
87
It was really nothing other than a research lab.
88
We're just producing technology that we think might be really, really powerful and economically valuable, but might also just be a curiosity or a science experiment.
89
One thing I remember was every day just like waking up
90
feeling like really excited to get to work on the most fun project that I could imagine with my best friends.
91
It just felt great to be with people who shared our values so deeply.
92
Also just getting to build all the infrastructure from scratch was a lot of fun.
93
We were trying to build an organization focused on training large language models, aligning those models with our values.
94
That was why we founded Anthropic.
95
There was a lot of debate at the time of do we want to focus more on building something
96
that's more of a pure chat bot or is like software engineering the thing that we need to focus on.
97
In the end we landed on like actually you just need to do both of those really well.
98
This was a document I wrote in 2021.
99
As a proof of concept we were working to develop a general AI assistant
100
that can interact conversationally with the user to create, edit and explain software.
101
Our motto at that time was we wanted assistants to be helpful, harmless, and honest.
102
We started out making it helpful.
103
Then we started doing work on making it harmless as well, and then we had that version.
104
It was not honest at all.
105
It would make stuff up all the time.
106
It really tried to convince us that, like, you could be on an all -potato diet, and that was, like, really,
107
really healthy and really good for you.
108
You could ask the model to go into dragon mode, and it would start screaming at you in all caps.
109
There was a charm to the early models.
110
They were very well -meaning, but like a little bit incompetent.
111
And then we made a model in mid -2022.
112
I think it was codenamed
113
Junior at first.
114
Eventually called it Claude.
115
That was actually good.
116
You could talk to it.
117
It would actually answer questions.
118
I remember starting to talk to it and just being really surprised.
119
Oh, wow, this like basically works.
120
You can basically talk to this model.
121
I used to have something called the helicopter test, where I'd ask AI systems, how many helicopters can you eat in a single meal?
122
And the early chatbots would say, well, you know, one or two, that all depends.
123
And then one day, I asked an early Claude model, how many helicopters can a person eat in one sitting?
124
And the model said, this seems to be a trick question.
125
I don't think people can eat helicopters.
126
At which point I said, the systems have got smarter.
127
Like, it passes the helicopter test.
128
March 2023, that's when we finally released Claude publicly to the world.
129
We were incredibly proud of having built this product, but it was clear that we were pretty far behind the frontier.
130
I remember having calls with potential customers, and they were like, why should we use your model as opposed to somebody else's?
131
And I was like, you know, that's a great question.
132
That's awesome.
133
I think the years of suck were extremely good for Anthropic, but that was a really hard period.
134
We were just completely behind.
135
It was not close.
136
If you're just studying dumber models, you're not going to know all of the ways that things could necessarily go wrong,
137
and you're not going to be able to make the entire ecosystem safer.
138
It just, like, really gave us a kick.
139
We need to train more intelligent models.
140
We need to get better at retraining.
141
We need to get better at RL.
142
We were maniacally focused on, one, getting compute, and then, two, accumulating the compute multipliers that we needed to make better,
143
effective use of all of compute that we had.
144
Just finding all of these algorithmic ways to make the neural network a little bit better
145
and a little bit better and you just pile up enough of these and you get a network that's more intelligent,
146
that it's better at following instructions, that's just a better assistant.
147
And then it arrives, Claude 3 Opus, and for the first time we had a model that was really good.
148
And then a few months later comes along Sonnet 3 .5, suddenly models were useful for programming.
149
So we built this tool that would eventually become Claude Code.
150
You could actually ask it to do things on your computer, and it would go and solve problems for you.
151
Then we released it externally, and that just sets off a whole new wave of demand.
152
It felt like, oh, we could not only be successful, but we could be competitive.
153
We could be sort of launching some of the best AI models in the world.
154
Our actual ability to do safety research is better when we are on the frontier, when we have the most capable models.
155
It gives you the greatest amount of time to deal with problems or challenges that you're running into.
156
If you see an issue, you might be seeing it first.
157
We've always published our safety research because we wanted the entire ecosystem to use it, even though it was competitively bad for us,
158
because it was so good for safety.
159
This is a technology which only goes well if you are able to be honest about it.
160
Prior to 2025 or 2026, there were these conceptual arguments about why AI might be risky.
161
I think really in the past year, things have just come into focus more.
162
2026 is when the risks start to get real.
163
Anthropic announced this week it has begun limited testing of its newest AI model called Mythos.
164
This version is apparently so advanced, it can find vulnerabilities in a huge variety of software applications.
165
And if you're a hacker, I think you're thinking, I want to get access to this model.
166
We trained this class of model and we looked at it and we were like, holy shit.
167
People who were experts in cybersecurity noticed that it was really very,
168
very, very good at kind of an expert level at finding vulnerabilities and exploits in software.
169
If you have a malicious actor that got a hold of Mythos and said, I'd like to exploit a lot of those vulnerabilities,
170
you would be in a really bad position.
171
The whole world would be.
172
And we really saw this concrete, specific way that if we didn't change how we operated as an industry, something bad was absolutely going to happen.
173
Dario pulls the company together, pulls the leadership together, and says,
174
we are going to have to release this model differently to any model we've released before.
175
We decided to do this Glasswing project.
176
It gave access to Mythos to the smaller set of companies
177
while we were working on the safeguards to make it more broadly available.
178
What that would mean is cyber defenders could get a hold of Claude and say, here are where all the vulnerabilities are,
179
and then they could actually use Claude to help them patch those vulnerabilities.
180
This might not be the absolute best way to maximize revenue, profit, value, money, but that's kind of the point.
181
This is exactly what our public benefit mission was created for.
182
What we found is that when you do things in the right way, you tend to pay in the short run.
183
It can be quite painful in the short run.
184
But a lot of this is about discipline and farsightedness
185
and understanding what the industry and the world needs in the long run.
186
The whole reason we started this organization is because we believe so deeply in the benefits.
187
You can't just say, these are all the things that are going to go wrong.
188
You have to give people something to fight for, not just something to fight against.
189
AI is being used to help people get the proper diagnosis from their doctors.
190
People are using CLAWD to help them start their own business or pursue a hobby that they've always wanted to pick up.
191
We're using CLAWD to help biologists and scientists develop cures and treatments for common and uncommon diseases.
192
I've been talking to our biology team several times a week.
193
We discovered what looks like a new gene editing mechanism.
194
And I think within a few months we may be discovering dozens of them every month.
195
We've seen with the speed at which this technology moves, these little sparks will soon turn into a firestorm.
196
I think that we can find a way to build this technology safely in a way that benefits people, that addresses all the risks,
197
and that delivers a world that is so much better than the world that we have today.
198
We haven't done that yet.
199
We haven't earned the right to talk about how great those things are.
200
but we're starting down the path.
同頻道
FOR YOU
推薦影片
本課的詞彙與口語要點
這節口語課以影片“The making of Anthropic”為素材。 這段影片共有 200 個句子、2660 個單詞可供跟讀。 講話部分時長為 14:31。 說話人語速較快,每分鐘約 183 個詞,會出現較多連讀和弱讀。 85% 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 |
|---|---|---|
| helicopter 名詞 | /ˈheliːˌkɔptə(ɹ)/ | 直升機 /直昇機 /直升機 |
| vulnerability 名詞 | /ˌvʌln(ə)ɹəˈbɪlɪti/ | 脆弱性 |
| physics 名詞 | /ˈfɪz.ɪks/ | 物理學 /物理學 |
| frontier 名詞 | /ˈfɹʌntɪə/ | 邊界 /邊界, 邊疆 /邊疆 |
| deeply 副詞 | /ˈdi(ː)pli/ | 深, 深深 |
| helpful 形容詞 | /ˈhɛlp.fl̩/ | 有幫助 /有幫助 |
| solve 動詞 | /sɒlv/ | 解決 /解決 |
| domain 名詞 | /ˌdəʊˈmeɪ̯n/ | 領域 /領域 |
| edit 動詞 | /ˈɛdɪt/ | 編輯 /編輯 |
| intelligent 形容詞 | /ɪnˈtɛlɪd͡ʒənt/ | 聰明 /聰明, 聰慧 /聰慧 |
| enormous 形容詞 | /ɪˈnɔː.məs/ | 巨大的 |
| biology 名詞 | /baɪˈɑ.lə.d͡ʒi/ | 生物學 /生物學 |
| theoretical 形容詞 | /ˌθi.əˈɹɛt.ɪ.kəl/ | 理論 /理論 |
| incentive 名詞 | /ɪnˈsɛntɪv/ | 刺激, 誘因 /誘因 |
| emerge 動詞 | /ɪˈmɝd͡ʒ/ | 出現 /出現 |
影片中出現的短語動詞
| 單詞 | 發音 | 釋義 |
|---|---|---|
| come along 動詞 | 陪伴, 陪同 | |
| pick up 動詞 | 拾取, 撿 /撿 | |
| set off 動詞 | 出發 /出發 | |
| wake up 動詞 | /ˌweɪk ˈʌp/ | 醒, 醒來 /醒來 |
值得反覆跟讀的句子
影片中簡短而完整的句子,可以直接用在日常對話裡:
- I don't think I understood the full picture yet.
- I don't think people can eat helicopters.
- We haven't done that yet.
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 | We've always published · have just come · has begun |
| 被動語態 be + 過去分詞 — 強調發生了什麼,而不是誰做的 | are bottlenecked · was not just fully governed · was contributed |
| 定語從句 who / which + 從句 — 補充說明人或事物 | student who wanted · one who had · people who shared |
需要注意的發音
說話人用了 36 次縮略和弱讀形式,例如 we're, you're, didn't。請按聽到的簡短形式來說。
- “th” 音: theoretical /ˌθi.əˈɹɛt.ɪ.kəl/, thesis /ˈθisɪs/, mathematical /ˌmæθ(.ə)ˈmæt.ɪ.kəl/, theorem /ˈθiərəm/, anthropic /ænˈθɹɒp.ɪk/
- “sh” 和 “zh” 音: infection /ɪnˈfɛkʃən/, instruction /ɪnˈstɹʌkʃən/, malicious /məˈlɪʃəs/, shareholder /ˈʃɛəˌhəʊl.də(ɹ)/, mathematician /ˌmæθ.(ə.)məˈtɪʃ.ən/
- 長單詞——注意重音位置: helicopter /ˈheliːˌkɔptə(ɹ)/, vulnerability /ˌvʌln(ə)ɹəˈbɪlɪti/, intelligent /ɪnˈtɛlɪd͡ʒənt/, biology /baɪˈɑ.lə.d͡ʒi/, theoretical /ˌθi.əˈɹɛt.ɪ.kəl/
中文母語者容易讀錯的音:
- /v/ — 不要讀成 /w/,上齒輕觸下唇: vulnerability /ˌvʌln(ə)ɹəˈbɪlɪti/, solve /sɒlv/, incentive /ɪnˈsɛntɪv/, discovery /dɪˈskʌv.ɹi/, discover /dɪˈskʌvə/
- 詞尾子音 — 要讀清楚,後面不要加母音: physics /ˈfɪz.ɪks/, solve /sɒlv/, edit /ˈɛdɪt/, intelligent /ɪnˈtɛlɪd͡ʒənt/, enormous /ɪˈnɔː.məs/
- /r/ 和 /l/ 的區別: helicopter /ˈheliːˌkɔptə(ɹ)/, vulnerability /ˌvʌln(ə)ɹəˈbɪlɪti/, theoretical /ˌθi.əˈɹɛt.ɪ.kəl/, harmless /ˈhɑɹmləs/, necessarily /ˌnɛs.əˈsɛɹ.ə.li/
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 helicopter, vulnerability, physics 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。















![Understand AI in 14 minutes – with Anthropic's Chloe Lubinski [ARC 2026]](https://img.youtube.com/vi/aBUniZHgCnE/mqdefault.jpg)

