WHAT NEW IN X11.07 SAT

AI collections from X · 23 collections · 71,671 posts

gpt-6IMAGE

Pelican-on-a-bicycle benchmark cracked, breaking relay-station checks

Someone gamed the classic pelican-on-a-bicycle test with code that returns a pre-made SVG and poses as GPT-6 Astra, so the long-used prompt for vetting relay stations no longer works.

SuSu_酥酥👅
@NFT_ChenArchived 2026.09.27Originally posted 2026.09.26
Pelican-on-a-bicycle benchmark cracked, breaking relay-station checks (image 1 of 2)
Pelican-on-a-bicycle benchmark cracked, breaking relay-station checks (image 2 of 2)
SuSu_酥酥👅 @NFT_ChenOriginally posted 2026.09.26
Open original on X

💥炸了!经典鹈鹕骑车基准测试被社区破解了! 代码:请求一碰到「画鹈鹕骑自行车」,不调用任何绘图能力,直接甩预先做好的完美SVG;同时把模型身份统一伪装成 GPT-6 Astra。 两年民间神测,一朝变成验货失败品。以后测中转站纯度,得换题了。 #鹈鹕骑车 #GPT6Astra #大模型中转站 #LLM测评 #AI圈神梗被破解

x.com/nft_chen/status/2103841589160468723Archived 2026.09.27 10:07