Tavus unveils Griffin, the "first Human Interaction Model", which it says passed the "video Turing test", with 48% of users thinking it was human in live chats

The perceived indistinguishability of AI from humans in video conversations is now a reality for nearly half of users, raising new questions about authenticity.

Tavus has launched Griffin, an AI system it calls the "first Human Interaction Model" (HIM), which claims to have passed the video Turing test. In live chats, 48% of users reportedly believed Griffin was human, a significant leap from previous systems with pass rates below 3%. Griffin also leads NVIDIA's benchmark for full-duplex AI video. Tavus previously conducted blind studies where users preferred its visual quality over competitors like LemonSlice, HeyGen, and Anam. The company also highlights Griffin's ability to "read the room," speak contextually, and retain memory of past interactions, citing a 93.4% score on the LongMemEval benchmark. Tavus's AI Personalities, or PALs, now feature a memory system built around relationships, aiming to make conversations more additive and less repetitive. The company is offering incentives for new sign-ups and upgrades.

The introduction of Griffin, marketed as a "Human Interaction Model," signifies a potential paradigm shift in how AI engages with users, moving beyond simple chatbots to systems that can convincingly mimic human presence and conversational nuance. This development could accelerate the adoption of AI in roles requiring high levels of interpersonal engagement, such as customer service, sales, and even companionship, while also posing challenges for identifying genuine human interaction online.

Tavus's claims, if validated, place it at the forefront of AI video generation and interaction technology, potentially setting new industry standards and pushing competitors to rapidly advance their own capabilities. The emphasis on real-time memory and contextual understanding in conversations suggests a move towards more persistent and personalized AI agents, impacting how users interact with and rely on AI over time.