Tavus has introduced Griffin AI, a model designed to make video conversations with artificial intelligence feel more like talking to another person. Its demonstrations show an assistant that laughs, reacts with surprise and interrupts rather than simply waiting for each question.
The company describes Griffin as a human interaction model, or HIM. It says 48% of participants in its video Turing test believed they were speaking with a real person.
However, that result represents a company-reported finding. The supplied announcement does not provide enough methodological detail to independently assess its significance.
Conversations Move Beyond Question-and-Answer Exchanges
Griffin’s demonstrations focus on the small behaviours that shape everyday conversation.
Alongside spoken answers, the model produces facial expressions and gestures that respond to the exchange. For example, it can laugh or appear surprised while interacting with someone on a video call.
Tavus also demonstrated the assistant interrupting a speaker and responding when someone interrupted it. These exchanges aim to reproduce the overlapping speech and changing pace of human conversation.
However, displaying an emotional expression does not establish that a model experiences the corresponding emotion.
Gestures Add to the Visual Realism
The demonstrations also show Griffin waving, touching its face and tucking hair behind an ear.
According to the presentation, its appearance remains visually consistent during these movements. That matters because awkward hand movements or distorted facial features can make generated video appear less convincing.
Instead of relying solely on a realistic face, Griffin combines speech, movement and conversational timing. Together, those elements create the impression of a responsive video-call partner.
Nevertheless, selected demonstrations cannot establish how consistently the model performs across longer conversations or unfamiliar situations.
Tavus Proposes Tutoring and Practical Assistance
Beyond casual conversation, Tavus showed Griffin helping with activities such as solving a Rubik’s Cube and acting as a practice partner.
Other examples involved following timing and directions during a project. The company consequently sees potential uses in tutoring, collaborative tasks and ongoing conversational support.
These examples illustrate the roles Tavus wants the technology to fill. However, they do not establish its reliability across every educational or practical task.
For now, Griffin-Lite is available as a research preview to a selected group of testers. Meanwhile, Tavus plans to offer a more powerful version more widely in the future.
The supplied announcement does not specify a date for that broader release.
Humanlike Video Raises Questions About Disclosure
Tavus describes Griffin as the first AI to pass a video Turing test. Its reported 48% result, however, needs context about participant selection, conversation length and test conditions.
Without those details, the figure alone cannot establish that the model is broadly indistinguishable from a human.
The demonstrations also prompted concerns about impersonation and transparency. Some social media users questioned whether similar systems could stand in for applicants during remote job interviews.
Others said they could still identify the presenter as artificial. Nevertheless, the discussion highlighted a practical issue: viewers may need clear disclosure when a realistic video participant is AI-generated.
As Griffin AI moves beyond limited testing, its conversational realism will remain only one part of the assessment. Consistency, usefulness and transparency will also shape how people respond to it.
0 responses to “Griffin AI Brings Interruptions and Lifelike Reactions to Video Chats”