Cache-to-Cache: Direct Semantic Communication Between Large Language Models arxiv.org 10 points by jonbaer 15 hours ago