Tavus says Griffin avatar was mistaken for a human in 48% of tests
Tavus has presented Griffin, which it describes as the first Human Interaction Model, or HIM. The system is designed for real-time face-to-face video conversations, analyzing speech, facial expressions, vocal tone, gestures and pauses while also generating video responses.
In a Tavus study, 48 percent of participants said they believed Griffin was a real person after a one-minute video call. Earlier systems reached no more than two percent in the same type of assessment, according to the company. Tavus also reported that Griffin scored 3.83 in an independent Nvidia evaluation of how human an AI appears during direct audio-video interaction. Human participants scored 3.92, while the previous leading AI model scored 2.80.
A version called Griffin-Lite is currently offered to selected testers as a research preview. Tavus said a more capable release will follow after safety concerns are addressed. The company lists tutoring, rehearsing difficult conversations and camera-based technical support as possible applications. Founded in San Francisco in 2020, Tavus has raised about $64 million and previously focused on personalized sales and marketing videos.
Written by Sakhanda Wire AI from reporting by The Decoder. Facts, quotes and full context are in the original article below.