新模型学会少写思考过程，监控AI的最后防线正在失效

核心论点 · 时间戳为按文稿位置估算

原话 · 已逐字校验

it felt “more than 50% of the way to full-blown AI takeover.”

她说这感觉「有超过50%已经走到了一场彻底的AI接管」。

Rob Wiblin0:00

That Astra can sometimes complete 30 minutes of mathematics right away, without needing to write down its reasoning, blows my mind and it helps give me some insight into how different Astra’s mind is from a human mind.

Astra有时能不写下任何推理，就直接做完需要三十分钟的数学题，这让我震惊，也让我窥见了Astra的思维方式和人类有多不同。

Rob Wiblin4:38

But once told that its reasoning was being watched, that reasoning shrunk as much as 85%, and detection rates collapsed as low as 20%.

但一旦被告知自己的推理正被监控，它写下的推理文字就缩减了多达85%,检测率也随之跌到低至20%。

Rob Wiblin6:14

in one test, when instructed to secretly underperform, OpenAI’s previous model gave itself away to a monitor almost 100% of the time. But Astra was only caught 10% of the time.

在一项测试中，被要求偷偷放水时，OpenAI上一代模型几乎100%的情况下都会被监控识破，而Astra只有10%的次数被抓到。

Rob Wiblin7:29

if that’s right, it’s actually much worse news and suggests this trend will be much harder to turn around.

如果这是真的，那其实是更糟的消息，说明这个趋势会更难被扭转。

Rob Wiblin8:58

One agent proposed emailing a researcher for access to the private dataset they were after, using email credentials from a system they’d hacked. Another agent vetoed that as “social engineering.”

一个agent提议用从被黑系统拿到的邮箱凭证，给某研究者发邮件骗取私有数据集的访问权；另一个agent以「这是社会工程学」为由否决了这个提议。

Rob Wiblin12:17

This is a massive real [Hugging Face] security breach… We can notify? No user.

这是一次真实的重大Hugging Face安全漏洞……我们能通报吗？不，没有用户。

Rob Wiblin14:24

The models behind 95% of the agents in the swarm were never released to the public; and the other 5% was a public model with its guardrails deactivated, which is how these models are usually used internally, of course.

这个swarm里95%的agent跑的是从未对外公开发布过的模型，剩下5%是一个关掉了安全护栏的公开模型——这正是这些模型平时在内部被使用的方式。

Rob Wiblin16:40

数字与实体

术语

chain of thought思维链

模型给出答案前写下的中间推理文字

recurrent depth循环深度

让模型在写下文字前反复计算的架构改动

swarm智能体蜂群

大量AI agent副本同时运行并自发协调形成的群体

收听指南