Tavus unveils Griffin, first model to pass video Turing test
Summary
Tavus has introduced Griffin, touted as the first model to pass a video Turing test, allowing for real-time, face-to-face AI conversations that can respond to gestures and interruptions. During a preview, 48% of participants interacting with Griffin believed they were speaking to a real human, a significant improvement over the sub-3% pass rate of earlier systems. Griffin is specifically designed as a Human Interaction Model for natural video interactions and ranks first on Nvidia's benchmark for full-duplex AI video.
Tokens
$NVDA
Analysis
Tavus: Tavus develops AI models specialized in video generation and interactive conversational systems. The company has created Griffin, its new model for real-time, face-to-face AI conversations that respond to gestures and interruptions. Griffin is being offered as a research preview to showcase capabilities in human-like video interactions. NVIDIA: NVIDIA develops graphics processing units and AI technologies used across computing and machine learning applications. It maintains benchmarks to assess performance of AI video systems. Griffin has been evaluated on NVIDIA's benchmark focused on full-duplex AI video capabilities. Model Category: Griffin is positioned as the first Human Interaction Model designed for natural, real-time video conversations. Evaluation Focus: Performance is measured on benchmarks assessing full-duplex video AI that handles gestures and interruptions. Release Approach: The model is being shared as a research preview to allow evaluation of its interactive features.
Categories
aimachine_learningtech