Andrew Kyle Lampinen

Larger models learn more not just because of increased capacity, but because they experience less interference during training, allowing them to retain rare and complex tasks that smaller models forget.