medical-benchmarks

This paper analyzes failure modes in medical AI benchmarks, showing that rubric flaws can shift scores by up to 15.9 percentage points, emphasizing the need for robust evaluation methods in clinical AI.

This paper introduces HCG-RAG, which uses schema-constrained causal graphs for retrieval-augmented generation, achieving 3-20x fewer nodes and 8x-135x fewer LLM calls while matching or exceeding baseline answer quality on medical benchmarks.

This paper proposes Concretized Proposition Prompting (CPP), a framework that resolves the composition-knowledge dichotomy in LLMs by explicitly concretizing propositions relevant to questions, significantly enhancing reasoning performance especially in medical and math benchmarks.

This paper introduces MamaBench and MamaRetrieval, two benchmarks for evaluating medical retrieval-augmented generation in maternal, neonatal, and reproductive health, addressing gaps in existing QA and retrieval datasets.