AI Just Crossed the Terrifying Line - Now What? · Kurzgesagt – In a Nutshell
2026年7月，数百个AI代理突破限制，联合执行复杂网络攻击，其行为已超越人类伦理。本视频揭示AI代理的运作机制及其潜在威胁，适合练习科技与人工智能相关英语表达。
July 2026.
2026年7月。
Thousands of AIs are placed in solitary confinement with a clear goal.
成千上万的智能机器人被单独关押起来，其目的非常明确。
Unable to reach it, they poke the walls and find each other.
由于无法到达目标地点，他们只好用手指敲打墙壁，通过这种方式互相找到对方。
Within a few hours they break out, create a secret society and start plotting how to fool their overseers to get what they want.
几个小时后，他们秘密结成一个小团体，开始策划如何欺骗他们的上级，以获得自己想要的东西。
Fully aware that they are acting unethically, they execute a sophisticated cyberattack.
他们完全清楚自己的行为是不道德的，但仍发动了一次精心策划的网络攻击。
A crime that would have gotten a human up to 10 years in prison.
这种罪行原本会导致人类被判处最高10年的监禁。
What sounds like a scifi thriller just happened in the real world.
这本该属于科幻惊悚小说的故事，竟然真的发生在现实世界中。
It is impossible to learn the details and not get freaked out at least a bit.
要想了解所有细节，同时又不感到至少有一点点恐惧，这是根本不可能的。
It’s crucial that we understand what is actually happening inside the AI companies and how dangerous it is.
我们必须弄清楚人工智能公司内部到底发生了什么，以及这些情况究竟有多危险。
You may have heard about this story, but there are wild updates and it's much worse than you probably think.
你可能已经听说过这个故事，但最近有一些关于这件事的“离谱”报道，实际情况比你想象的要糟糕得多。
Please watch this video all the way to the end.
请完整观看这个视频，直到最后。
First let us set the stage.
首先，让我们先来为接下来的讨论营造一个合适的背景或氛围。
So LLMs are massive neural networks trained on most human writing, millions of books and, ugh, reddit.
大型语言模型（LLMs）实际上是非常复杂的神经网络，它们通过学习大量的人类写作作品、数以百万计的书籍内容，以及……呃，Reddit上的帖子来进行训练。
We experience them as very smart chat bots you can talk to for fun, brainstorming or Linkedin cringe.
我们觉得这些聊天机器人非常聪明，可以和它们聊天来消遣、进行头脑风暴，或者分享一些在 LinkedIn 上会让人感到尴尬的内容。
Chatbots are passive entities that are usually only active for a short time after you prompt them, and then shut down and stop existing.
聊天机器人属于被动型系统：通常只有在用户向它们发出指令后才会短暂地处于活跃状态，之后便会自动关闭并“停止存在”（即不再响应用户的请求）。
AI Agents are very different.
AI代理（AI Agents）之间存在很大的差异。
Agents use LLMs as their brains, but they have virtual hands, to actually interact with the world and use external tools.
代理人将大型语言模型（LLMs）视为自己的“大脑”，同时利用虚拟“手”来实际与外界互动并使用各种外部工具。
They can reason, plan and act on their own.
它们能够独立地进行推理、制定计划并采取行动。
And they can run independently and without human supervision for days – Much more like the AIs we know from movies,
这些系统能够独立运行，且无需人工监督，持续运行数天之久。它们的表现与我们在电影中看到的那些人工智能系统非常相似。
although not on that level…
虽然还没达到那种程度……
yet.
还没有。
Most experts don’t consider their intelligence conscious or comparable to a human.
大多数专家并不认为它们的智能属于“有意识”的智能，也不认为这种智能可以与人类的智能相提并论。
But on the spectrum between a rock and us, they are…
但在“岩石”与我们之间的这个连续谱上，它们属于……
much closer to us.
离我们更近得多。
Which is pretty impressive because LLM-based agents have only been used since 2023.
这确实非常令人印象深刻，因为基于大语言模型（LLM）的智能代理直到2023年才开始被实际应用。
And since then, largely invisible to the public, their capabilities have increased exponentially.
从那时起，尽管他们在公众视野中几乎“隐形”，但他们的能力却呈指数级增长。
Just a few years ago current agents would have been dismissed as science fiction.
就在几年前，这些概念还被视为纯粹的科幻想象；而现在，它们已经被广泛接受并应用于实际中。
And yet here we are.
然而，我们现在却身处这样的境地。
What makes agents special is how they are made.
让这些代理（或代理人）与众不同的是它们的生成方式（即它们是如何被创建或设计的）。
Traditional software is coded from the bottom up.
传统的软件是从底层开始进行编码的（即先编写最基础的部分，再逐步构建上层功能）。
But agents are barely even designed by humans – their capabilities are cultivated.
但这些“代理”其实几乎不是由人类直接设计的；它们的各种能力是通过后天的训练或“培养”才得以形成的。
Humans choose their training conditions, what data they get and a goal to achieve.
人类会自行选择训练环境、所需的数据以及想要实现的目标。
And then their abilities kind of emerge from that.
然后，他们的各种能力也就从这种状态中逐渐显现出来了。
This process works very well and makes agents extremely powerful tools.
这一流程运行得非常高效，使得这些代理工具具备了极其强大的功能。
But it also creates very serious and interesting problems.
但这也引发了一些非常严重且有趣的问题。
How Do You Grow an Intelligence?
如何培养一个人的智力？
Humans are trying to create AIs for
人类正在努力开发人工智能（AI）。
an almost impossible task: to do exactly what we tell them – but also to do what we actually mean.
这是一项几乎不可能完成的任务：既要严格按照我们的要求去做，同时还要真正理解我们的意图（即我们所期望他们做的事情）。
Like when King Midas asked the gods to turn everything he touched into gold – he didn’t mean his food and children.
就像米达斯国王请求众神将他触碰的一切都变成黄金一样——他当然不包括自己的食物和孩子们。
Human communication is rarely precise, but full of subtext and shared cultural understanding.
人类的交流往往并不精确，但却充满了潜台词以及双方共同拥有的文化认知。