Reproduce “Human Evaluation with Automatic Scores: Separating Sampling and Estimation Gains” 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
Yang Liu, Ran Zou, Zhengtao Gui, Chao Xue
The authors compare score-based corrections and label allocations with a proportional, score-stratified human mean on fixed, fully labeled evaluation pools. They replay sampling under matched label budgets, measuring mean squared error against each pool’s full human average. After ten-bin stratification, even a retrospective oracle correction reduces error by only 1.00–5.27%; practical gains vary by pool and decision.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.