signal极客公园2026-10-04
Nodding, Interrupting, Blushing: Half the People Can't Tell She's an AI Girl
Tavus released the Griffin model on October 1, claiming it as the world's first Human Interaction Model. In a one-minute video call test, 26 of 54 participants thought they were chatting with a real person, a 48% pass rate compared to 2.4% from the previous system. Griffin uses a full-duplex, video-to-video architecture, generating 720p video in 320-millisecond blocks with an average reaction delay of 0.43 seconds.
- for who
- AI digital human developers and interaction designers building real-time conversational avatars.
- why now
- Tavus released Griffin on October 1, nearly fooling half of testers in live video calls.
- what changes
- The benchmark shifts from making faces look realistic to making AI converse and respond like a human, with silence and interruptions as natural cues.
- to do
- Evaluate Griffin's full-duplex capabilities and compare it to cascaded pipelines for real-time conversational video use cases.
key points
- Griffin achieves 48% human pass rate vs 2.4% for previous system
- Full-duplex video-to-video model with 0.43s average reaction delay
- Uses Tavec audio codec and Self-Forcing training for real-time generation
#ai digital human#tavus#griffin#full duplex#real-time video
score
score 8 out of 10. 0-10: how dense the facts are, multiplied by how much you can do with them after reading. 8+ means the topic's evidence bar is met: benchmarks and availability for a new model, amount and investors for a funding round, revenue figures for a solo-money story. Below 5 an item does not enter the digest. A press release scores 3 or less, a reprint loses 2, anything older than 14 days loses 1, a headline that misleads loses 3.
read the source