Should We Still Pretrain Encoders with Masked Language Modeling? - Explained Simply | ArXiv Explained