New AI Model Nearly Indistinguishable from Humans in Video Conversations

According to Tavus, an AI video company based in San Francisco, its newly released model Griffin-Lite deceived 26 out of 54 participants (48%) into believing they were speaking with a real person during one-minute video calls. Under the same testing conditions, only 2.4% of participants made this judgment with the previous generation system. Tavus states that this is the first real-time video conversation model to pass the 'video Turing test,' defining it as a 'human interaction model.' Unlike traditional voice assistants, Griffin-Lite responds in real time even when both parties speak simultaneously, mimicking natural human chat by either agreeing or interrupting.