Djamé Seddah

Standardized benchmarks have become the dominant metric for measuring progress in large language models, yet their validity is increasingly compromised by data contamination and the unclear relationship between benchmark scores and genuine language understanding. We introduce Gaperon, a suite of...