Fork the consciousness, or download the project and create your own.

Wei Xu

Xi'an Jiaotong University; ShanghaiTech University

Builder The claim is about what the hardware does, and the philosophy is left open

Position on AI consciousness

Xu's team treats awareness as a measurable capability layer that is distinct from language-modeling progress, a separation their own data shows through near-identical Elo models differing by over 24 points on self-awareness.

Wei Xu is the senior author of AwarenessBench, the September 2026 benchmark from Tsinghua University, the Shanghai Qi Zhi Institute, and partner institutions that measures 15 awareness functions across 18 language models and 299,628 evaluation turns. The benchmark’s construction adapts existing datasets where coverage exists and designs new tasks where it does not, drawing on GPQA-Diamond, Humanity’s Last Exam, ToMi, Big-Bench, and the 2024 self-recognition tasks of Laine and colleagues.

On this site’s taxonomy the profile is Builder, for the resource built and maintained rather than a philosophical stance. The team’s published disclaimers carry the consciousness-relevant weight. Awareness is presented as a practical proxy for consciousness, not a measurement of experience, with an explicit instruction that contemporary models are not conscious and a stated caution against using the benchmark for cultivating machine consciousness. The scientific finding with the largest consequence for the AI consciousness debate is the distinct-capability result, that progress in language modeling or reasoning does not necessarily translate into improved cognition, which grounds the behavioral-evidence limits this site tracks from the theory side.

Known for. Senior author of AwarenessBench, the awareness benchmark spanning metacognition, self-awareness, social awareness, and situational awareness in language models

Coverage on this site

Related researchers