Should We Still Pretrain Encoders with Masked Language Modeling? Paper • 2507.00994 • Published Jul 1 • 77 • 9