Our work “Should We Still Pretrain Encoders with Masked Language Modeling?” is accepted at ICLR 2026!