1 paper
Sheng Jin, Minghao Liu, Yunze Xiao +24
Knowledge benchmarks for LLMs face three issues: scaling-driven designs that do not operationalize disciplinary representativeness; flat-payment annotation that permits lazy consen…