AI Model Griffin-Lite Surpasses Previous Systems in Human Interaction
- Tavus’ AI model, Griffin-Lite, convinced 48% of participants it was human during a one-minute video call.
- Griffin-Lite scored first on Nvidia’s VideoFDB benchmark with a generation score of 3.83 out of 5.
- The previous Tavus system had only a single participant out of forty-one believe it was human.
- Participants were not informed they might be speaking to an AI during the test conducted by Tavus.
- Griffin-Lite is currently available only to select trusted testers as Tavus develops safety measures.
Tavus’ Griffin-Lite has demonstrated significant advancements in AI-human interaction, achieving nearly half the participants believing they were speaking with a real person during trials. This marks a substantial improvement from their earlier systems and highlights the potential for more natural AI communication models in the future.
