AI Model Griffin Achieves Near-Human Performance in Job Interview Simulations
A new artificial intelligence model developed by Tavus AI has demonstrated a remarkable ability to mimic human interaction, with nearly half of participants in a recent test mistaking the AI for a human interviewer.
A new artificial intelligence model named Griffin, developed by the Bay Area startup Tavus AI, has shown a significant capacity to emulate human conversation. In a recent job interview simulation, 48% of the 54 participants believed they were interacting with a real person, not an AI.
Tavus AI claims this result indicates that Griffin can pass the Turing Test, a benchmark designed to assess a machine's ability to exhibit intelligent behavior indistinguishable from that of a human. Participants who were convinced the interviewer was human expressed an average of 79% confidence in their assessment. Conversely, those who identified it as a bot were 81% confident.
The Griffin model reportedly generates every frame of its output in real-time, starting from a single reference image. This includes not only the avatar's facial movements, arms, and fingers but also environmental details such as chair movement, shadows, and background elements. The avatars are available in male and female appearances and exhibit natural-looking breathing and blinking.
"Humans are evolutionarily designed to communicate face to face. We speak as much through our words as we do our expressions, tone, gestures and timing. So much of human communication is non-verbal. It’s an art, a dance," stated Hassaan Raza, CEO and co-founder of Tavus, in the company's announcement.
Raza also highlighted that Griffin addresses issues seen in previous AI-generated avatars, such as inappropriate emotional responses like smiling when receiving bad news. The model is designed to react in real-time, capable of interrupting users, adapting to new topics, and maintaining conversational flow even when interrupted.
Demonstrations of the Griffin model showcased its ability to respond to visual cues, such as guiding a participant through solving a Rubik's Cube. It was also shown successfully playing a game of Simon Says. Tavus AI suggested potential applications for its models, including use in educational settings as tutors for students.