Tavus's video avatar fooled nearly half of test subjects in a one-minute call
San Francisco startup Tavus released real-time video avatar Griffin; its own study found 48% mistook it for a human, worth watching for independent verification.
ImportanceLocalEvidenceE2 unreplicatedWrite-upQuick
After a one-minute video call, 48% of test subjects believed Tavus's Griffin avatar was a real person.
According to The Decoder on October 1, the San Francisco startup introduced Griffin, which it calls the first "Human Interaction Model," processing speech, facial expressions, tone and gestures in real time while generating video. A Tavus-run study produced the 48% figure; previous systems maxed out at 2%.
Tavus also cites an independent Nvidia test it describes: Griffin scored 3.83 on how human it feels, versus 3.92 for actual humans and 2.80 for the previous best AI. The test's sample size and design are not stated in the report.
Only a preview called Griffin-Lite is available to select testers; a fuller version awaits resolved safety concerns.