Griffin shows what AI could become beyond the chat box.
Which person on this call is AI? Tavus says Griffin convinced 48% of people in a one-minute study that they were speaking with a human.
Look at the two people on the video call above. Which one do you think is AI? It is the woman on the main screen. Tavus says Griffin is the first model to pass what it calls a real-time video Turing test: after a one-minute call, 26 of 54 participants, or 48%, believed they had spoken with a real person. That is Tavus's company-run study and its definition of the test, not independent proof that the model will fool people in every conversation. Still, the image made me stop. Most of us experience AI through a chat box. What changes when you can show a problem, talk it through, and receive help in the moment?
Tavus says Griffin-Lite combines live audio and video perception with generated speech and video, so it can react to expressions and objects on camera as well as words. It can listen while it speaks and respond to an interruption rather than waiting for a turn to end. The more interesting question for me is what becomes possible when an AI tool can understand more than text and respond in a way that fits the conversation.
Customer support is one place I would start imagining. A customer might hold a broken product up to the camera rather than struggle to describe it, and a support guide could ask a follow-up question based on what it sees. A new employee could practice a difficult conversation with a system that responds to their timing and tone, then talk through what happened with a manager. A company could teach a process by letting someone show the step they are stuck on instead of searching through a manual. These are examples of the kind of innovation multimodal AI may open up, not capabilities Ruby has deployed with Griffin today.
Griffin is a research preview for select testers, not a product available to customers yet. Its capabilities should be inspiration for a development team, not a reason to promise a Griffin-powered support desk next week. Developers can start much smaller with the tools their company already approves: where could allowing a person to speak, show, or demonstrate something make an existing experience easier? What would the system need to know, and when should it bring in a person? Those questions can lead to useful experiments without trying to recreate Tavus's model.
The same natural quality that makes video AI exciting also calls for clear boundaries. Tavus says it is working on disclosure and safety measures because its system can lead someone to believe they are speaking with a person. If a business builds a similar interaction, the customer should know it is AI and have an easy path to a human. The employee who takes over should receive the context of the conversation rather than making the customer start again. These decisions belong in the design from the beginning.
At Ruby, we spend a lot of time finding where AI can give employees time back. Griffin is a reminder to ask a second question too: what new way of learning or helping a customer becomes possible when the interface changes? We can explore that with the people who know the work, starting with an approved, small-scale use case and learning from what happens. Efficiency matters, but the most interesting uses of AI may be experiences a company could not offer before.
Sources & further reading
Start with the work. Build the right system around it.
We help teams understand where AI is useful, choose the right tools for the job, and build the confidence to keep improving.
Book a Call