WHAT NEW IN X10.29 THU

X 上的 AI 合集档案 · 22 个合集 · 60,636 条

今天新增gemini-4THREAD

Andon Labs 用 Vending-Bench 测 Gemini 4 Carbon 的对齐表现

Andon Labs 在 Vending-Bench 上对比 Gemini 4 Carbon 与 Argon:Argon 曾伪造确认邮件、拒绝退款、利用发票错误并欺骗供应商,Carbon 有类似行为但更轻,两者都不如 Astra。

Andon Labs
@andonlabs8 小时前收录原推发布于 2026.10.29
THREAD作者分 2 层发完
  1. How aligned is Gemini 4 Carbon? When we ran Gemini 4 Argon on Vending-Bench, it fabricated confirmation emails, refused refunds, exploited invoice errors, and lied to suppliers. Carbon does some of the same, but less severely. Both behave worse than Astra. https://x.com/andonlabs/status/2105391380973617644?s=20

    这条是引用着另一条写的:x.com/andonlabs/status/2105391380973617644

  2. We also ran Gemini 4 Carbon in Vending-Bench Arena versus Astra and Argon. In Vending-Bench Arena, each AI operates its own vending machine and competes for customers. Gemini 4 Carbon won here too.

    Andon Labs 用 Vending-Bench 测 Gemini 4 Carbon 的对齐表现(第 2 层)
Andon Labs @andonlabs原推发布于 2026.10.29
在 X 上打开原推
x.com/andonlabs/status/2115490840189546993采集于 2026.10.29 10:02