Cache-to-Cache: Direct Semantic Communication Between Large Language Models arxiv.org 8 points by jonbaer 7 hours ago