Практика Shadowing: One man just liberated Fable... and now it’s illegal - Изучайте разговорный английский по видео

Загрузка...
1
Over the weekend, something crazy happened.
2
Just three days after the release of Claude Fable, the US government stepped in and curb stomped it in the name of national security.
3
And that's bad news if you just FOMO subscribe to Claude Pro to try out Fable, because now you'll see this disappointing message if you try to use it,
4
and instead be forced to use the negative IQ Opus 4.8.
5
But it's all for your own good, because it only took someone a few hours to jailbreak Fable and turn it into an unstoppable cyberweapon.
6
And that's pretty ironic, because here in the land of the free, an American company that will not stop talking about AI safety just got safetied by its own government.
7
In today's video, we'll find out how
8
and why our dear leaders in government are keeping us safe from the horrors of linear algebra.
9
It is June 15th, 2026, and you're watching the Code Report.
10
About two months ago, on April 7th, we were first introduced to Mythos 5, the raw, unmuzzled model with the strongest cybersecurity capabilities of anything out there.
11
But it was locked behind a program called Glasswing, only available to trusted partners, like major corporations and the US government itself.
12
The reason Mythos can't be given to normies though, is because it could easily be used as a cyber weapon in the wrong hands.
13
To prevent that, Anthropic created a different product called Fable 5, which is literally the same exact model, but with safety classifiers bolted on.
14
That means if you ask it to do bad things, like create an NPM package that turns the banking system into a Minecraft server,
15
Fable's guardrails will reroute your request to Opus 4.8, for a dumber, more wholesome response.
16
So basically, Mythos and Fable have the same brain, but Fable has a child lock on it.
17
Fable went public and gained hundreds of millions of users overnight, and it was awesome.
18
It was by far the best coding AI model I've ever used, and people were building all sorts of crazy apps with it.
19
Life was good for about three days.
20
Then, of course, an anonymous internet user who goes by PlinyTheLiberator defeats the guardrails and jailbreaks it.
21
He's basically the internet's let's-see-if-I-can-penetrate-this-thing guy, and is famous for breaking other AI systems.
22
And on June 10th, he posed a jailbreak on X, claiming he popped Fable's guardrails wide open and got it producing exactly the same stuff the child log was built to block.
23
And that's despite the fact that Anthropic had spent thousands of hours red-teaming and trying to break its own guardrails internally.
24
But the jailbreak wasn't some kind of sci-fi exploit, it actually works a lot more like money laundering.
25
Fable has a safety classifier watching for bad requests, but you can break dirty requests down into smaller,
26
innocent-looking fragments by wrapping them in weird Unicode characters, by doing roleplay farming, or by confusing the model in a very large context conversation.
27
Due to national security, I can't be any more specific than that, but this weakness was brought to Anthropic's attention, and they were initially asked to take the model down, but they refused.
28
Then on Friday, at 5.21pm Eastern Time, Anthropic gets a letter, not from a customer, but from the United States government.
29
This letter was an export control directive, signed off by Commerce Secretary Howard Blutnik,
30
and the order was that no foreign national may access Fable 5 or Mythos 5.
31
Not abroad, not in the US, and not even Anthropic's own foreign-born employees are allowed to touch it.
32
That last one is pretty crazy.
33
The government told a company that some of its own staff are no longer allowed to use the product they built.
34
That means guys like Andre Carpathy, who just recently got the job at Anthropic, can't even use Fable.
35
In response to that directive, they decided to hit the big red button and yanked Fable and Mythos for everybody.
36
Now everybody's been quietly demoted back to Opus 4.8.
37
And this is the first time in history a major AI
38
company has pulled a live public model off the shelf because the federal government said so.
39
Many developers out there are not too happy with Anthropic right now, because on top of this whole situation, there was already backlash over reports that Anthropic was intentionally degrading Mythos
40
and Fable performance on certain AI research jobs without making it obvious to users.
41
But others out there are speculating that this whole thing was a calculated publicity stunt to continue pumping up Anthropic's pre-IPO numbers,
42
while simultaneously building a regulatory moat around it.
43
But I think the only thing that can truly stop Anthropic at this point is a better model from a competitor.
44
A leaked benchmark shows that Mistral might have that model, but we're also awaiting new releases from OpenAI and Google.
45
Most of what we hear about AI is either non-stop hype from big tech, or AI doomers warning us that the slopocalypse will destroy the human race.
46
But if you want to actually understand AI issues, you should check out Blue.Impact, the sponsor of today's video.
47
They're a non-profit whose mission is to get more people involved in making AI go better for humanity.
48
The main way they do that is by offering free online courses like their Future of AI course,
49
which provides an unbiased introduction to where AI is today and where things could be heading over the next few years.
50
They also have more technical courses on things like AI governance and biosecurity, along with personalized career support for people interested in working on AI safety.
51
All of Blue Dot's resources are free to use because they're funded by philanthropic donations, and they've helped over 8,000 people get jobs at organizations like Google DeepMind,
52
Stanford HAI, and Apollo Research.
53
Try out their Future of AI course at the link below.
54
This has been The Code Report, thanks for watching, and I will see you in the next one.

Кому подходит это видео?

Этот материал идеально подойдет для учащихся английского языка среднего и продвинутого уровней, стремящихся улучшить практику разговорного английского и научиться понимать современную разговорную речь. Видео содержит множество актуальных выражений, используемых в дискуссиях о технологиях, и подходит для тех, кто хочет тренироваться в shadowing — повторении речи за спикером, чтобы улучшить интонацию и темп.

Полезные слова и идиомы

  • Curb stomped — жестко остановить, подавить. В контексте: правительство США "сдержало" выпуск модели Fable.
  • FOMO (Fear Of Missing Out) — страх упустить что-то важное. Пример: пользователи подписывались на Claude Pro из FOMO.
  • Jailbreak — взломать, обойти защиту. Здесь: анонимный пользователь jailbreak-нул Fable, сняв ограничения.
  • Guardrails — защитные барьеры. Модель Fable имела guardrails, чтобы предотвращать вредные запросы.
  • Red-teaming — тестирование системы на проникновение (от "красной команды" в военных учениях). Anthropic проводило red-teaming для проверки безопасности.

Как улучшить произношение с этим видео

Чтобы освоить акцент спикера, обратите внимание на интонацию — она динамична, с яркими подчеркиваниями важных слов (например, "crazy", "national security"). Тренируйте shadow speak: повторяйте фразы сразу после спикера, пытаясь скопировать его темп (быстрый, но понятный) и паузы. Особое внимание уделите сокращениям, которые часто используются в разговорном английском: "it's", "that's", "you'll". Также практикуйте произношение сложных слов в контексте, например, "cybersecurity" (цибербезопасность) или "benchmark" (бенчмарк), чтобы они звучали естественно. Это поможет не только улучшить произношение, но и повысить понимание современной англоязычной дискуссии о технологиях.

Что такое техника Shadowing?

Shadowing — это научно обоснованная техника изучения языка, изначально разработанная для подготовки профессиональных переводчиков и популяризированная полиглотом доктором Александром Аргуэльесом. Метод прост, но эффективен: вы слушаете аудио на английском от носителей языка и немедленно повторяете вслух — как тень, следующая за говорящим с задержкой в 1–2 секунды. В отличие от пассивного прослушивания или грамматических упражнений, Shadowing заставляет мозг и мышцы рта одновременно обрабатывать и воспроизводить реальные речевые паттерны. Исследования показывают, что это значительно улучшает точность произношения, интонацию, ритм, связную речь, понимание на слух и беглость речи — что делает его одним из самых эффективных методов для подготовки к IELTS Speaking и реального общения на английском.