Skip to content

Translation in progress Some entries still include text in the original language.

EventPast event

Scale AI 与 Elorian 发布 HSS 直觉视觉推理基准

1 reports1 reporting sourcesLatest report

Start here

AI overview

Scale AI 与 Elorian 联合发布 Humanity's Sixth Sense(HSS)基准,用于评测直觉视觉推理,包含 288 张图片和 234 段视频(共 17.6 小时)上的 522 道开放式任务。人类参与者准确率 93.1%,最强模型 GPT-6-astra 仅 53.6%,中位模型 30.9%;移除图像或视频后 GPT-6-astra 降至 6.6%。评测覆盖 8 家厂商的 25 个多模态模型。

Generated by AI from reports · Updated 3 days ago

Progress timeline

Oct 7
  1. Scale AI Research Blog
    Scale AI 与 Elorian 发布 HSS 基准,测出多模态模型直觉视觉推理远低于人类

    Scale AI 与 Elorian 联合发布 Humanity's Sixth Sense(HSS)基准,用于评测直觉视觉推理,包含 288 张图片和 234 段视频(共 17.6 小时)上的 522 道开放式任务。人类参与者准确率 93.1%,最强模型 GPT-6-astra 仅 53.6%,中位模型 30.9%;移除图像或视频后 GPT-6-astra 降至 6.6%。评测覆盖 8 家厂商的 25 个多模态模型,模型平均每题消耗 4046 个推理 token,提高推理强度在部分子领域反而下降,如 GPT-6-astra 在 retrodiction 上从 high 到 xhigh 掉 14 分。8573 次失败中 94% 源于感知或潜在推断,仅 5% 源于逻辑错误,最主要原因是漏掉关键视觉线索(21%)和误认对象、人物或角色(20%)。

Related discussion

No discussion links have been collected for this event yet.

Attention trend

New sources per day

6 days total1 ·Maximum1(Oct 7)·Today0

  • Oct 7New1 sources (1 outlets,0 accounts)
  • Oct 8New0 sources (0 outlets,0 accounts)
  • Oct 9New0 sources (0 outlets,0 accounts)
  • Oct 10New0 sources (0 outlets,0 accounts)
  • Oct 11New0 sources (0 outlets,0 accounts)
  • Oct 12New0 sources (0 outlets,0 accounts)

Hourly attention

Current attention43·Peak in the comparable period44(Oct 8 02:00)·Change over the comparable past 24 hours 0%

0204060Oct8Oct9Oct11Oct12