[–] jedharris 1y ago ↗ "Our results show that simple algorithmic improvements can enable significantly more data-efficient pre-training in a compute-rich future."
1 comment
[ 3.0 ms ] story [ 20.9 ms ] thread