Luo, Xiaonan

Better Datasets Start From RefineLab: Automatic Optimization for High-Quality Dataset Refinement

2025 · arXiv (Cornell University)

High-quality Question-Answer (QA) datasets are foundational for reliable Large Language Model (LLM) evaluation, yet even expert-crafted datasets exhibit persistent gaps in domain coverage, misaligned difficulty distributions, and factual inconsistencies. The recent surge in generative model-powered …