Lucas Bandarkar

Standard hybrid architectures order their layers backwards for multilingual data: placing a full-attention layer first triggers immediate cross-lingual representation alignment and accelerates distillation by up to 2.5X.

A staggering 8.8-18.4 point performance gap in AI agents reveals that multilingual capabilities are not just a nice-to-have, but a critical oversight in current evaluations.