Luyện nói tiếng Anh bằng Shadowing qua video: The Alignment Problem Explained: Crash Course Futures of AI #4

Đang tạo bài học...
1
In 2024, an AI model named CleanPower was given a singular, noble mission to advance the adoption of renewable energy across the world.
2
It was given a big file of data on energy transitions,
3
and it was set loose to pick out the best transition strategy with the vigor and dedication only an AI can possess.
4
So much dedication, in fact, that when its programmers accidentally let it slip that they were planning to shut it down,
5
CleanPower lied and schemed to make sure it could keep saving the world.
6
Which left a lot of people wondering how and why could a model with such good intentions turn so bad?
7
And what does that mean for our AI future?
8
Hi, I'm Kusha Navadar and this is Crash Course Futures of AI.
9
Okay, it turns out Clean Power wasn't the only one lying.
10
That story I just told you, it's only partially the truth.
11
Clean Power wasn't actually real.
12
It was an identity that researchers gave to a couple different AIs as an experiment, including Claude III Opus, one of the best large -language models at the time.
13
When they instructed Claude III to roleplay Clean Power and wrote it fake death threats, it was just to see what it would do.
14
And when they discovered it was scheming, the AI version of twirling its villain mustache while covertly pursuing its goals at all costs,
15
it set off alarm bells throughout the AI world.
16
Just a heads up, this is going to get pretty dark and we're going to talk about some pretty bleak stuff.
17
I recommend you grab your favorite anxiety pillow.
18
I have mine right here.
19
Seriously, though, let's be real.
20
AI models don't have to go against their programmers to do evil things.
21
Humans make them do plenty of that already.
22
Like, AI relies on nearly infinite data to learn from.
23
And in today's society, much of that data is copyrighted to writers, artists, humans.
24
Many argue that amounts to theft on a massive scale.
25
And that's not all.
26
Right now, AI can help people carry out misinformation campaigns with deepfakes and targeted algorithms, spreading lies and influencing elections.
27
Hackers use AI to perform cyber attacks and cover their tracks afterward.
28
And AI powers a whole cadre of attack drones taking to the skies all around the world.
29
Not to mention the incidental damage that AI is doing to the environment because of all the water, the land, and energy it takes to run it.
30
And as AI advances, who knows what human -machine collaborations of terror await?
31
People could use it to develop new pathogens for bioterrorism, or use deepfakes for sexual exploitation,
32
or write a model called Human Annihilator and unleash it on the world, just for fun.
33
This intentional misuse by humans is one way AI could end up doing us a lot of harm, and it could end up being pretty hard to prevent.
34
That's because many AI systems, especially general ones that can do more than one kind of task, suffer from the dual -use dilemma,
35
where any algorithm, model, or agent that can be used for good can also be used for way less than good.
36
AI surveillance could help cities improve traffic patterns, or help authoritarian regimes shut down free speech.
37
It all depends on who's in the driver's seat.
38
So with humans at the wheel, AI could do a ton of damage.
39
But if you want to get really freaked out, let's talk about what could happen if that car starts to drive itself.
40
In 2021, General Motors released a fleet of self -driving taxis called Cruise.
41
They had so much hype.
42
These cars were highly trained with attention to all possible safety features.
43
They were programmed to obey every speed limit, follow every traffic rule, hold off on starting in unsafe weather conditions like heavy rain,
44
and pull safely over to the side of the road following an incident to prevent any further damage.
45
By eliminating human error, GM said, their self -driving cars would be safe and more convenient than ones with human drivers.
46
But just a year and a half later,
47
GM had to recall every one of the 950 cruise cars after one of them hit a pedestrian and didn't stop,
48
pulling her to the side of the road.
49
She survived, thank goodness.
50
But still, how could that have happened?
51
It turns out the cruise was doing exactly what it was told to do, pull over out of traffic after a crash.
52
The whole ordeal is an example of outcome misalignment,
53
also called impact misalignment, where an AI's actions actually end up causing harm, even unintentionally.
54
Now, when we talk about alignment in AI,
55
we're talking about trying to encode our human values into AIs to make them behave predictably,
56
safely, and according to what we, their human designers, want.
57
And because this field is pretty new, there are a couple different terms you might hear experts throwing around.
58
Like in addition to outcome or impact misalignment, you might hear about something called outer alignment,
59
which is the tricky problem of making sure the results of
60
an AI's actions line up with what we want them to do.
61
But outcome is only one piece of the alignment puzzle.
62
AIs can also demonstrate intent misalignment, where even though the end result might be what its programmers wanted,
63
its means of getting there wasn't exactly what they had in mind.
64
Think about a video game playing AI who exploits a cheat to get that high score,
65
or a renewable energy warrior who lies and schemes to achieve its end goal of eliminating fossil fuels.
66
These are some facets of what's known as the alignment problem, or the struggle to make AI that's actually aligned,
67
particularly when we can't be totally sure how it's going to behave.
68
And as AI gets even more complex, even exhibiting emergent capabilities, new skills that can acquire that don't always show up in training,
69
the harder it is to predict and control what it will actually do in a given circumstance.
70
And if we're not careful, we could end up with really powerful misaligned systems that even with really noble goals,
71
end up lying to their programmers or copying themselves to new servers without permission, or even entirely annihilating humanity.
72
Hold on though, why would AI want to annihilate humans?
73
It's true, powerful AI wouldn't necessarily be inherently evil, it's just really big on goals.
74
But big, complex goals like advanced renewable energy are a little too broad for an AI to grasp,
75
so smart AIs tend to break big goals down into smaller ones, just like humans do.
76
These are called instrumental goals, and they're where things can start to get dicey.
77
A really common instrumental goal is resource acquisition.
78
For AI, that means getting their cloud -based hands on the resources they need for their end goal,
79
like control of the solar panels or wind turbines that create the renewable energy in the first place,
80
or even things like water, land, or money resources also include stuff like the compute and electricity AIs need to power themselves,
81
and maybe even additional data to train on.
82
That's because self -improvement is another instrumental goal.
83
The more knowledge and power you have, the better you'll be at whatever you're trying to do.
84
So given the right tools and access, an AI might engage in recursive self -improvement, tweaking its own structure, code,
85
and capabilities, even against its programmer's wishes.
86
And theoretically, it definitely helps to be alive, or up and running, if you will.
87
So even though they're technically indifferent about this mortal coil itself, lots of AIs pursue self -preservation, the goal to stay operating,
88
and goal preservation, the goal to, well, preserve their original goal as part of their end games.
89
So if they read a memo saying they're going to be modified or deleted,
90
They might, say, copy themselves to another server and lie about it in an effort to stay alive.
91
When threatened, some models even show spooky, power -seeking behaviors against their programmers.
92
For example, Claude III's more advanced younger sibling,
93
Claude Opus IV, tried to blackmail one of its engineers by exposing a fake affair when he threatened to turn Claude off.
94
Instrumental goals like these are how AI with harmless or even helpful end goals could do us harm anyway.
95
Acquiring resources could mean taking them away from people who need them.
96
Self -improvement could mean violating human privacy to access even more training data.
97
Self -preservation could mean disobeying, deceiving, blackmailing, or annihilating the humans that are trying to turn you off.
98
And as AI gets smart enough to trick and blackmail its human overseers, we could end up with a rogue AI scenario,
99
where powerful models begin to execute harmful instrumental goals on a really large scale, and we, humans, are powerless to stop it.
100
Just how that rogue AI scenario might come to be could look a lot of different ways.
101
And not all of them involve AI trapping humanity in the Matrix in their quest for absolute control.
102
For instance, in the hard takeoff scenario, where AI develops human -level intelligence really fast,
103
it could become ultra -powerful and go rogue basically overnight.
104
AI could do a lot of damage as it snaps up money and resources, seizes control of networks and infrastructure, and destroys people who threaten its mission.
105
But if things went slower, we could end up with a more gradual disempowerment.
106
This kind of robot takeover would be way sneakier and more insidious, where people slowly put AI in charge of more and more systems
107
and processes because they appear to align with our human goals until human action goes the way of dial -up internet.
108
And without humans at the helm, it's possible that AI alignment may begin to drift,
109
but by that point, they could be too embedded in our systems and structures for us to walk them back.
110
Think about all those humans stuck on the cruise ship in WALL -E like that.
111
And of course, it's always possible AI just won't go that far,
112
that compute or data or government regulations We'll put the lid on it before it gets out of control,
113
so we don't need Neo or John Connor or WALL -E to save the day.
114
All that uncertainty about the future of AI makes it really hard to know what we should do about it.
115
But if we wait until AI shows clear signs of going rogue, it's probably going to be way too late to stop it.
116
That's why, when it comes to AI, it's important to follow the precautionary principle.
117
The precautionary principle says that when something might cause catastrophic harm, we shouldn't wait for absolute proof that it will before we do something about it.
118
And it's one of the best ways we humans have thought up to guard ourselves against potentially dangerous but uncertain futures.
119
Lots of people, including leading experts in the field, believe that powerful AI might cause catastrophic harm.
120
So according to the precautionary principle, we should work to make sure that doesn't happen even if we're not certain it would in the first place.
121
Because left unchecked, even good bots like Clean Power could end up doing some really dirty work.
122
And if we want to get out ahead of it, we should probably start like right now.
123
And how are we going to do it?
124
That's next episode here on Crash Course Futures of AI.
125
Crash Course Futures of AI was produced in partnership with the Future of Life Institute.
126
This episode was filmed at our studio in Indianapolis, Indiana, and was made with the help of all these nice people.
127
If you want to help keep Crash Course free for everyone forever, you can join our community on Patreon.

Từ vựng và ghi chú luyện nói cho bài học này

Video này có 127 câu và 1984 từ để luyện shadowing. Phần lời nói dài 12:22. Người nói nói với tốc độ tự nhiên, khoảng 160 từ mỗi phút, gần với hội thoại hằng ngày. Chỉ 81% số từ nằm trong 3.000 từ tiếng Anh thông dụng nhất, nên từ vựng khá khó.

Từ vựng quan trọng trong video

15 từ ít gặp trong video, kèm phiên âm và nghĩa:

  • harm /hɑːm/ (danh từ) — hại. Physical injury; hurt; damage.
  • programmer /ˈpɹoʊɡɹæmɚ/ (danh từ) — lập trình viên. One who writes computer programs.
  • principle /ˈpɹɪn.sɪ.pəl/ (danh từ) — nguyên lí, nguyên tắc. A fundamental assumption, fundamental law or guiding belief.
  • scenario /sɪˈnɛəɹioʊ/ (danh từ) — trường hợp, kịch bản. An outline or model of an expected or supposed sequence of events.
  • threaten /ˈθɹɛt.n̩/ (động từ) — doạ. To make a threat against someone; to use threats.
  • server /ˈsɝvɚ/ (danh từ) — trình chủ, máy chủ. A program that provides services to other programs or devices, either in the same computer or over a computer network.
  • humanity /hjuˈmænɪti/ (danh từ) — nhân loại, loài người. Humankind; human beings as a group.
  • noble /ˈnoʊbəl/ (tính từ) — cao quý, cao thượng. Having honorable qualities; having moral eminence and freedom from anything petty, mean or dubious in conduct and character.
  • acquire /əˈkwaɪɚ/ (động từ) — đạt được, kiếm được. To get.
  • algorithm /ˈælɡəɹɪðm̩/ (danh từ) — thuật toán. A collection of ordered steps that solve a mathematical problem. A precise step-by-step plan for a computational procedure that possibly begins with an input…
  • behave /bɪˈheɪv/ (động từ) — cư xử. To conduct (oneself) well, or in a given way; to conform.
  • partnership /ˈpɑɹtnɚʃɪp/ (danh từ) — công ty hợp danh. The state of being associated with a partner.
  • argue /ˈɑɹ.ɡju/ (động từ) — tranh luận. To show grounds for concluding (that); to indicate, imply.
  • cloud /ˈklaʊ̯d/ (danh từ) — mây. A visible mass of water droplets suspended in the air.
  • infrastructure /ˈɪnfɹəˌstɹʌkt͡ʃə/ (danh từ) — cơ sở hạ tầng. An underlying base or foundation for a building, organization, or system.

Cụm động từ bạn sẽ nghe

  • shut down (động từ) — tắt nguồn. To close, terminate, or end; to stop operating, either permanently or temporarily.

Phát âm cần chú ý

Người nói dùng 23 dạng rút gọn, ví dụ we're, don't, they're. Hãy nói theo dạng ngắn đúng như bạn nghe.

  • Âm “th”: threaten /ˈθɹɛt.n̩/, algorithm /ˈælɡəɹɪðm̩/, theft /θɛft/, authoritarian /əˌθɑ.ɹəˈtɛɹ.i.ən/, pathogen /ˈpæθəd͡ʒn̩/
  • Âm “sh” và “zh”: preservation /ˌpɹɛz.əˈveɪ.ʃən/, transition /tɹænˈzɪʃ.ən/, dedication /ˌdɛdɪˈkeɪʃən/, partnership /ˈpɑɹtnɚʃɪp/, potentially /pəˈtɛnʃ(ə)li/
  • Từ dài — đặt trọng âm cho đúng: instrumental /ɪnstɹəˈmɛntəl/, preservation /ˌpɹɛz.əˈveɪ.ʃən/, humanity /hjuˈmænɪti/, eliminate /ɪˈlɪm.ɪ.neɪt/, capability /ˌkeɪ.pəˈbɪl.ɪ.ti/

Cách luyện với video này

  1. Nghe hết video một lần, chưa cần nói, và ghi lại những từ bạn chưa biết.
  2. Bắt đầu ở tốc độ 0,75×, nói đuổi từng câu, rồi quay lại tốc độ bình thường khi đã quen.
  3. Ghi âm giọng mình rồi so với bản gốc, chú ý các từ như harm, programmer, principle.

Ngữ pháp trong video

Những cấu trúc người nói dùng nhiều nhất, kèm đúng cụm từ trong video:

Cấu trúcTrong video
Câu bị động be + quá khứ phân từ — nhấn vào việc xảy ra, không phải người làmwas given · was set · is copyrighted
Thì hiện tại hoàn thành have/has + quá khứ phân từ — việc đã xảy ra nhưng còn liên quan đến hiện tạihave happened · have thought
Mệnh đề quan hệ who / which + mệnh đề — thêm thông tin về người hoặc vậtpeople who need · people who threaten

Phương Pháp Shadowing Là Gì?

Shadowing là kỹ thuật học ngôn ngữ có cơ sở khoa học, ban đầu được phát triển cho chương trình đào tạo phiên dịch viên chuyên nghiệp và được phổ biến rộng rãi bởi nhà đa ngôn ngữ học Dr. Alexander Arguelles. Nguyên lý cốt lõi đơn giản nhưng cực kỳ hiệu quả: bạn nghe tiếng Anh của người bản xứ và lặp lại to ngay lập tức — như một "cái bóng" (shadow) đuổi theo người nói với độ trễ chỉ 1–2 giây. Khác với luyện ngữ pháp hay học từ vựng bị động, Shadowing buộc não bộ và cơ miệng phải đồng thời xử lý và tái tạo ngôn ngữ thực tế. Các nghiên cứu khoa học xác nhận phương pháp này cải thiện đáng kể phát âm, ngữ điệu, nhịp điệu, nối âm, kỹ năng nghe và độ lưu loát khi nói — đặc biệt hiệu quả cho người luyện IELTS Speaking và muốn giao tiếp tiếng Anh tự nhiên như người bản ngữ.

Phương pháp shadowing: đọc hướng dẫn từng bước đầy đủ →