Back to Home
RadarAI··Papers & Tech

Elorian 发布 DecoyCount 基准,前沿模型视觉计数得分低于 50%

中文摘要

Elorian与Mercor发布DecoyCount基准,测试模型是真计数还是依赖视觉捷径,结果显示前沿模型视觉计数得分低于50%。

English Summary

Elorian and Mercor released DecoyCount to test if models truly count or use visual shortcuts; leading models scored below 50% in visual counting.

Original Excerpt

Latent Space 采访 Elorian CEO Andrew Dai,介绍其与 Mercor 联合发布的 DecoyCount 基准,测试模型是真正数数还是依赖视觉捷径。