Reproduce “Qworld: Question-Specific Evaluation Criteria for LLMs” 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

Shanghua Gao, Yuchang Su, Pengwei Sui, Curtis Ginder, Marinka Zitnik

Qworld generates question-specific evaluation criteria by expanding and consolidating scenarios, perspectives, and concrete checks before scoring answers with weighted criteria. On a 1,000-question HealthBench subset, it achieves 0.89 Coverage and 0.79 Uniqueness, versus EvalAgent’s 0.83 and 0.50. These semantic matches use an LLM judge, while human ratings sample criteria and do not validate every novel item.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.