Nearly half of test subjects mistook Tavus' AI video avatar for a real person on a one-minute call
decoder ·
Tavus has introduced Griffin, described as the first Human Interaction Model capable of real-time face-to-face conversations by processing speech, facial expressions, and gestures. In a study conducted by Tavus, 48 percent of participants believed Griffin was a real person after just one minute of video interaction, a significant improvement over previous systems that maxed out at two percent. An independent test by Nvidia further validated these results, showing Griffin scored 3.83 on human likeness compared to 3.92 for actual humans and 2.80 for the previous best AI model. A limited preview version named Griffin-Lite is currently available to select testers while safety concerns are addressed. The company identifies potential applications for this technology in tutoring, practicing difficult conversations, and camera-based tech support. Founded in 2020, the San Francisco-based startup has raised approximately $64 million since its inception. Originally focused on personalized AI videos for sales and marketing, Tavus has expanded its capabilities into live video conversations with digital personas.
decoder