Artificial intelligence is expanding past text-based chatbots with the introduction of a new model engineered for real-time video conversations. San Francisco-based AI firm Tavus has launched Griffin, which it characterizes as its inaugural Human Interaction Model (HIM).
Unlike standard chatbots, Griffin communicates via video. It listens to spoken queries, processes visual cues, and replies using speech, facial expressions, gestures, and body movements. The company states the system is engineered to make interactions feel significantly more natural and responsive.
Griffin Reacts in Real Time
According to Tavus, Griffin does not wait for a user to finish speaking. Instead, it operates continuously and tracks events as the chat unfolds in real time.
Consequently, if conditions change mid-conversation, the system can adapt immediately. It accommodates interruptions, shifts in tone, and cues derived from what appears on the screen. Tavus additionally notes that Griffin integrates video generation, speech processing, and conversational AI tools into a single package. The objective is to interpret signals like timing, tone, and facial expressions while sustaining a continuous conversation.
Tavus showcased the model in a demonstration video where Griffin answers questions while exhibiting human-like movements. The company asserts that these features can make it difficult for users to tell the AI apart from a living person during certain interactions.
Griffin Shows Human-Like Responses
Tavus shared several performance metrics for Griffin. The firm reported that 48% of individuals who engaged with the live model believed they were talking to an actual human.
Furthermore, Tavus stated that Griffin achieved a score of 3.83 out of 5 on NVIDIA’s VideoFDB benchmark. The human reference score stood at 3.92, whereas the company noted a score of 2.80 for the next-highest published system.
These statistics originate directly from Tavus and have not been independently verified in the original report.
Job Interviews Raise Concerns
This technology holds potential for sectors where natural dialogue is essential. Tavus highlighted applications such as customer support, tutoring, and interview practice.
The model’s capacity to interpret both speech and visual signals could also prove valuable for interactive simulations and training. Even so, employing an AI to pose as a candidate during a real job interview introduces major concerns regarding consent, disclosure, and the integrity of the evaluation process.
Certain employers currently ban candidates from utilizing generative AI during live interviews.
Griffin Remains in Testing
Tavus has not made Griffin publicly available without restriction. The enterprise reports it is actively developing safety protocols and disclosure measures ahead of a broader release.
At present, qualified testers can try Griffin-Lite through the company’s early testing program. Nevertheless, this development marks a clear progression toward AI systems that move beyond simply answering prompts to engage in lifelike, two-way video communication.
