Stories
US Startup Reveals AI Avatar That Fools 48 Percent in Job Interviews
- •Bay Area startup Tavus developed an advanced AI model named Griffin that convinced 48 percent of participants in a simulated job interview that they were speaking to a real human.
- •Griffin generates every pixel in real time using a single reference image, successfully replicating facial features, body movements, and surrounding background details.
- •Participants who mistook the AI avatar for a human expressed 79 percent confidence in their judgement, highlighting the technology's lifelike communication capabilities.
- •The AI is designed to mimic natural behaviors such as blinking, breathing, interrupting appropriately, and responding to changing visual information like solving a Rubik's Cube.
- •Tavus CEO Hassaan Raza emphasizes that the technology aims to bridge the gap in non-verbal communication, with potential future applications in education and interactive learning.
Key Questions Answered
Griffin is a new human interaction model developed by the Bay Area-based AI startup Tavus that generates every pixel of video frames in real time. During a simulated job interview test involving 54 participants, 48 percent of them believed they were interacting with a real human.
Griffin is designed to replicate natural human behaviors such as blinking, breathing, making gestures, changing facial expressions, and adjusting vocal tone. It can also handle interruptions appropriately, adapt to changing conversations, and respond to visual information like solving a Rubik's Cube.
Participants who mistakenly believed they were talking to a human were, on average, 79 percent confident in their judgment. Meanwhile, those who correctly identified the model as an artificial intelligence system were, on average, 81 percent confident in their decision.
