Oossa

Tavus AI avatar Griffin fools 48% of users in short video call

Tavus says 48% of study participants thought its avatar was human after a one-minute call. Griffin-Lite is available only to select research testers.

By Published by Oossa: 2 min read

visuals · Unsplash

Tavus has introduced Griffin, a system the San Francisco startup calls a “Human Interaction Model.” It is designed to hold face-to-face conversations in real time, taking in speech, facial expressions, tone, gestures and pauses while also generating video. The company says the system aims to handle more than spoken words alone, including the visual and vocal cues people use in conversation.

In a Tavus study, 48 percent of participants believed Griffin was a real person after a one-minute video call. Tavus says previous systems topped out at 2 percent. The Decoder does not provide the study’s sample size or explain how participants were recruited, so the result offers limited detail about how broadly it might apply.

What do the reported results show?

Tavus also cites an independent Nvidia test that measured how human an AI feels in direct audio-video conversation. Griffin scored 3.83 points, compared with 3.92 for actual humans and 2.80 for the previous best AI model. The Decoder does not name the test, give its scale or describe the settings, so the scores are difficult to interpret beyond that comparison.

These figures are company-reported results, not evidence that Griffin will seem human to everyone or perform equally well in other situations. The one-minute call study and the Nvidia test measure related but different things; The Decoder does not say they used the same participants or conditions.

Who can use Griffin?

Tavus says Griffin-Lite is available to select testers as a research preview. A more capable version is planned for later, after the company says safety concerns are addressed; it has not given a release date or described who will qualify. That means Griffin is not presented as a generally available product.

Tavus lists tutoring, practicing difficult conversations and camera-based tech support as possible uses. The startup, founded in 2020, began with personalized AI videos for sales and marketing before expanding into live conversations with digital personas. The Decoder does not report whether Griffin has been tested in those settings or how it handles errors and sensitive conversations.

Why it matters

For people considering AI tutoring or camera-based tech support, Griffin points to a possible way to have a more natural-feeling video interaction. But access is limited to select testers, and the reported results do not show whether it works well or safely in those real-world uses.

Was this article useful?
Share

Read next

Oossa · Newsletter

The week in AI, explained

Every Monday: the stories worth knowing, in plain language. Free, no spam.