1 paper
Zhuo Chen, Oriol Mayné i Comas, Zhuotao Jin +2
We present a universal theoretical framework for understanding long-context language modeling based on a bipartite mutual information scaling law that we rigorously verify in natur…