Cache-to-Cache: Direct Semantic Communication Between Large Language Models arxiv.org 4 points by jonbaer 4 hours ago