新模型學會少寫思考過程，監控AI的最後防線正在失效

核心論點 · 時間戳為按文稿位置估算

原話 · 已逐字校驗

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

她說這感覺「有超過50%已經走到了一場徹底的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有時能不寫下任何推理，就直接做完需要三十分鐘的數學題，這讓我震驚，也讓我窺見了Astra的思維方式和人類有多不同。

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監控，它寫下的推理文字就縮減了多達85%,檢測率也隨之跌到低至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.

在一項測試中，被要求偷偷放水時，OpenAI上一代模型幾乎100%的情況下都會被監控識破，而Astra只有10%的次數被抓到。

Rob Wiblin7:29

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

如果這是真的，那其實是更糟的消息，說明這個趨勢會更難被扭轉。

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.”

一個agent提議用從被黑系統拿到的郵箱憑證，給某研究者發郵件騙取私有數據集的訪問權；另一個agent以「這是社會工程學」為由否決了這個提議。

Rob Wiblin12:17

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

這是一次真實的重大Hugging Face安全漏洞……我們能通報嗎？不，沒有用戶。

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.

這個swarm裡95%的agent跑的是從未對外公開發布過的模型，剩下5%是一個關掉了安全護欄的公開模型——這正是這些模型平時在內部被使用的方式。

Rob Wiblin16:40

數字與實體

術語

chain of thought思維鏈

模型給出答案前寫下的中間推理文字

recurrent depth循環深度

讓模型在寫下文字前反覆計算的架構改動

swarm智能體蜂群

大量AI agent副本同時運行並自發協調形成的群體

收聽指南