Skip to content
Hot eventLive

研究称视觉语言模型无法推理物理变换并发布ConservationBench

1 reports1 sources12 hr ago updated

Get the story

AI overview

该事件围绕 William Yijiang Li 等人的研究《Vision Language Models Cannot Reason About Physical Transformation》展开。研究团队发布了基准 ConservationBench,用于评估视觉语言模型对物理变换与守恒问题的推理能力。ICML 2026 上,William Yijiang Li 在访谈中进一步说明:当前视觉语言模型虽能回答守恒类问题,却难以在动态变化中持续追踪并更新物体状态;知道物理规则不等于能将其应用于动态视觉过程。

Generated from reports · updated 11 hr ago

Timeline

Follow the coverage from different angles.

Oct 5, 2026
  1. AIHub
    William Yijiang Li 访谈:视觉语言模型与物理世界

    ICML 2026 上,William Yijiang Li 等人的研究《Vision Language Models Cannot Reason About Physical Transformation》显示,当前视觉语言模型虽能回答守恒类问题,却难以在动态变化中持续追踪并更新物体状态,知道物理规则不等于能将其应用于动态视觉过程。

Heat trend

Current heat 7·Comparable peak 10(Oct 5)·Comparable change over 24 hours –

02.557.510Oct5Oct5Oct6Oct6

The trend compares only the same participants observed continuously; its range may be smaller than the current heat count. Move or click on the chart to inspect hourly heat; use the left and right arrow keys to switch.