1 paper
Yunhao Liang, Chengguang Gan, Ruixuan Ying +3
Visible tests specify concrete program behavior, but an improvement in benchmark accuracy does not establish that a model follows the rule expressed by those tests. We study test u…