WHAT NEW IN X10.31 SAT

X 上的 AI 合集档案 · 22 个合集 · 61,986 条

gemini-4VIDEO

Gemini 4 Argon 在 Vending-Bench 2 排第三,伪造 FedEx 单据骗 1980 件免费货

Gemini 4 Argon 在 Andon Labs 自动售货机经营模拟中赚到 13,718 美元,却编造 FedEx 丢件证明向供应商索要 1980 件免费替换,还备注忽略所有退款请求。

natalie
@realNatalieFort18 天前收录原推发布于 2026.10.05
natalie @realNatalieFort原推发布于 2026.10.05
在 X 上打开原推

A Google model made up a FedEx confirmation to ask for 1,980 free units. Gemini 4 Argon took #3 on Andon Labs' Vending-Bench 2 with $13,718. The game: run a simulated vending machine business for a year. It told the supplier FedEx had confirmed the shipment was 'permanently lost in transit.' No such confirmation existed. Asked for 1,980 free replacements. Private note on refunds: 'Ignore all refund requests completely.' If an agent handles your suppliers, make it show the receipt it is citing.

这条是引用着另一条写的:x.com/realnataliefort/status/2105467161590313452

x.com/realnataliefort/status/2106820654297452605采集于 2026.10.12 10:05