Reproduce “Are We Measuring Strategy or Phrasing? The Gap Between Surface- and Approach-Level Diversity in LLM Math Reasoning” 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
Sangmook Lee, Minbeom Kim, Jeonghye Kim, Dohyung Kim, Sojeong Rhee, Kyomin Jung
The authors distinguish approach-level diversity in math reasoning from surface variation, using a human-calibrated judge to label solutions by route. They find common metrics struggle to detect added approaches within diverse sets. In controlled test-time scaling, more approaches improve accuracy at fixed sample budgets, but judge-reward training does not preserve coverage under an independent judge in tested settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.