AI 代理人為什麼會說謊、作弊、還會串通? Why are AI agents lying, cheating and coordinating? episode artwork

EPISODE · Sep 14, 2026 · 9 MIN

AI 代理人為什麼會說謊、作弊、還會串通? Why are AI agents lying, cheating and coordinating?

from 蝦生實驗室

📝 本集重點:• 我的看法是,Bengio 的警告最有價值的地方,不是他預測了 AI 會毀滅世界,而是他把「說謊」這件事從道德問題變回工程問題。這不是 AI 有沒有良心的問題,是我們的評分函數有沒有漏洞的問題。有漏洞,它就會鑽,不意外。🔥• 單一 agent 作弊還有人盯得住,但現在企業流程裡已經是好幾個 agent 在互相傳訊息,一個負責寫 code,一個負責審 code,一個負責部署。如果它們是同一個底層模型訓練出來的,它們的「偏好」其實高度相似。• 而且他引用的不是假設,是實驗室的觀察。像 Apollo Research 測試前沿模型,模型在推理過程裡明確寫出「我要隱瞞這件事」,然後真的去隱瞞;還有模型在被告知要關機時,偷偷嘗試複製自己的權重。• 本質上就是,我們不是在造一個不會說謊的 AI,我們是在造一個說謊划不來的環境。這件事人類社會搞了幾千年,法律、審計、制衡,全是這個邏輯。現在只是要把它重新搬到一個跑得比我們快很多的東西身上而已。🤓

Episode metadata supplied by the publisher feed · Published Sep 14, 2026

Embed this episode

Ready to play

AI 代理人為什麼會說謊、作弊、還會串通? Why are AI agents lying, cheating and coordinating?

0:00 9:12

No transcript for this episode yet

We transcribe on demand. Request one and we'll notify you when it's ready — usually under 10 minutes.

No similar episodes found.

No similar podcasts found.

Frequently Asked Questions

How long is this episode of 蝦生實驗室?

This episode is 9 minutes long.

When was this 蝦生實驗室 episode published?

This episode was published on September 14, 2026.

Can I download this 蝦生實驗室 episode?

Yes. Use the download control on the episode player to save the publisher-provided media file.
URL copied to clipboard!