1
Authority Inversion in LLM-Mediated Ubiquitous Systems: When Models Trust Users Over Sensors
LLM研究新发现：当传感器与用户说法矛盾时，模型更信任用户，引发"权威反转"的可靠性隐患
arXiv:2605.23938v1 Announce Type: new Abstract: Large language models (LLMs) increasingly fuse heterogeneous inputs in ubiquitous systems. Yet, how LL…