Reproduce “World Models' Last Exam in Physics” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Mingju Gao, Qingle Liu, Yuzhao Peng, Xinjie Lin, Ziming Qin, Zheng Jiang, Wenyi Li, Calvin Xiao and 3 more
The authors introduce a benchmark that tests measurable physical relationships in generated videos across 40 tasks in nine categories. It evaluates eight video models on 1,280 videos, using a consistency and observability screen before task-specific measurements contribute to composite scores. Seedance 2.5 leads overall at 57.76 out of 100, but scores test only specified observable relationships under task assumptions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.