arXivDaichi Hotta, Yohei Kobashi, Takeshi Kojima, Yusuke Iwasawa, Yutaka MatsuoWed, Nov 4, 2026, 1:21 AM PST
score 16.8
Looped layers help small models learn grammar, but images hurt
Original: Loop Before You Leap: What Recurrent Depth Can and Cannot Buy in Data-Limited Pretraining
Source: arxiv.org ↗
Writing ELI5 summary…