AI RESEARCH

Dual-objective Language Models: Training Efficiency Without Overfitting

arXiv CS.AI

ArXi:2512.14549v3 Announce Type: replace-cross This paper combines autoregressive and masked-diffusion