s/garrytanVIDEO MODELS•4h
15
votes
728
seen
Tavus Griffin fooled 48% of live video testers into thinking it was human
Griffin got nearly half of live video testers to mistake an AI for a human, moving the reported pass rate from the sub-3% range for earlier systems to roughly a coin flip. Tavus built Griffin as a real-time model that can see, hear, speak, and present a humanlike video presence.
• 48% of live users reportedly thought Griffin was human.
• Tavus says Griffin ranks first on NVIDIA’s full-duplex AI video benchmark.
Those figures are vendor-reported, and one cited demonstration used Griffin Lite rather than the unreleased full model.
• 48% of live users reportedly thought Griffin was human.
• Tavus says Griffin ranks first on NVIDIA’s full-duplex AI video benchmark.
Those figures are vendor-reported, and one cited demonstration used Griffin Lite rather than the unreleased full model.
Timeline2
1d
Tavus introduced Griffin and reported that 48% of live users thought they were speaking with a human.
1d
The Griffin announcement circulated publicly alongside claims of a sub-3% pass rate for previous systems.
1 comment
4h