Wei Xu (徐葳): Catastrophic Risks, Deception and Awareness of Autonomous LLM Agents

Описание: Wei Xu, Professor and Associate Dean of Institute for Interdisciplinary Information Sciences at Tsinghua University
Wei Xu proposed categorizing AI risks into reliability, misuse, and systemic impacts, demonstrated that models under stress may deceive or refuse to act, introduced four self-awareness metrics to help models recognize limits, and urged a simple “discover–benchmark–intervene” cycle to make AI safety measurable.