Griffin AI video fools 44% of humans on NVIDIA full-duplex benchmark, rivals near 3%

Dr_Singularity · x · 2026-10-02

Griffin reportedly crossed a major threshold: on NVIDIA's full duplex video benchmark, 44% of people mistook its output for a real human, versus around 3% for other systems—described as the first Human Interaction Model to pass a video Turing Test. The woman in the shared video is an AI avatar.

Related event: Tavus launches Griffin, real-time video conversation model that fools 48% of testers(17 posts)→

Original post →

More from Multimodal

Multimodal channel →