China publishes ‘landmark paper’ on AI-to-AI technique that kicks human ‘bottleneck’ out of the loop and replaces us with an AI ‘modem’ — C2C brainwave direct connection achieves 150% boost in inference speed

清华大学团队提出 Cache-to-Cache(C2C)技术,让两个大模型直接交换内部缓存信息,跳过"把想法写成文字再传给对方"的环节。论文已被 ICLR 2026 接收,并附带开源代码。








