Tavus: SI model Griffin passes a video version of the Turing test, with 26 of 54 participants taking it for a live human
Prepared by Sintelligent with the help of SI. Publication time restored from the build log. . Updated: .
The line between a live interlocutor and SI on video has almost vanished: the machine can already look and behave like a human.
According to Tavus itself, after a one-minute video call 48% of study participants decided they had spoken with a human; the company's previous system scored 2.4% (1 of 41). Griffin works as a single video-to-video system: it perceives sound and image simultaneously and responds in real time with speech and facial expressions, nodding, changing tone on the fly and reacting to gestures. Participants were told beforehand that they were talking to a human; the lighter Griffin-Lite was tested, it is available only to select testers, and the company is delaying a public release until safety questions are resolved: by its account, the same properties make it possible to mislead people.
Sources 1