Tavus' Griffin: 48% mistook its AI video avatar for a human on a 1-minute call
The Decoder · rss · 2026-10-02
Tavus introduced Griffin, billed as the first "Human Interaction Model," which holds real-time video calls while reading facial expressions, tone of voice and gestures. In the company's study, 48% of participants believed Griffin was a real person after a one-minute call — up from a 2% ceiling for previous systems.
More from Multimodal
- Garry Tan: This Realtime AI Video Agent Is Indistinguishable From a Human — garrytan · 2026-10-02
- Black Forest Labs adds up-to-4K image generation, opens commercial weights inquiries — bfl_ai · 2026-10-02
- Black Forest Labs launches FLUX 3 Image: pixel-perfect edits, 4K, open weights soon — bfl_ai · 2026-10-02
- Qwen 2.1 image editing struggles with pose changes while 2511 shines — Axymeus · 2026-10-02
- OneTrainer LoRA recipe: Prodigy optimizer plus a 3-version merge trick — jiggydancer · 2026-10-02
- NoSpoon Studios offers free AI music videos and microdramas on its last public day — Kyrannio · 2026-10-02