Tavus study: 45% mistook AI for human in video call; 28% doubted real humans
Tavus, a Sequoia-backed AI company, unveiled Griffin, a real-time video-to-video model that can interrupt, laugh, and react like a human during conversations. In company studies, 45-48% of participants believed they were interacting with a real person. On NVIDIA's Video Full-Duplex Benchmark, Griffin Lite scored 3.83 out of 5, near the human benchmark of 3.92, and far ahead of the next best system at 2.80. The full model remains unreleased pending policy review.
IllustrationEditorial responsibility
- No named human review is recorded for this page.
- Reports are grouped by semantic similarity and deterministic rules. Language models may assist titles, summaries, translation and cross-source analysis; the page reads the event directly, while its address stays stable when the title changes.
- Summary covers the current reports
Cross-source coverage
Reporting timeline
Tavus unveils AI video model that can interrupt, laugh, and react like a human
Tavus has announced a new AI video model capable of interrupting, laughing, and reacting like a real person during conversations. In a 'video Turing test', 48% of participants reportedly believed they were interacting with a human, according to the company's announcement. The model represents a significant advancement in AI-driven video interaction, aiming to blur the line between human and machine communication. The announcement was made via a post on Polymarket, a prediction market platform, highlighting the model's ability to mimic natural human conversational behaviors such as interruptions and emotional reactions. The 48% figure indicates that nearly half of test participants were unable to distinguish the AI from a real person, suggesting a high level of realism in the model's responses.
Tavus launches Human Interaction Model Griffin; 48% of live testers thought it was human
Tavus, a Sequoia-backed team, has launched the first Human Interaction Model (HIM) named Griffin. According to a post by analyst rohanpaul_ai, 48% of people who interacted with the AI live on a video call believed they were speaking with a real human. The post claims this is the first AI to pass the Turing test on a video call. The analyst expressed enthusiasm for the technology's potential applications, noting hundreds of possible use cases. The post includes a video demonstrating the AI's natural interaction, including a moment where the AI says 'I'm AI'.
Read sourceTavus introduces Griffin video-to-video model, claims first to pass Turing test
Tavus has introduced Griffin, a research preview of its newest video-to-video model designed for real-time face-to-face conversations. Griffin learns conversational behavior by watching and listening, allowing it to interrupt mid-sentence or respond when someone enters the frame. Tavus claims Griffin is the first model to pass the Turing test, with 45% of 120 study participants mistaking its conversation for real human interaction. On NVIDIA's Video Full-Duplex Benchmark, Griffin Lite ranks first among all evaluated models, scoring 3.83 out of 5 compared to 3.92 for real humans. The full Griffin model remains unreleased while Tavus reviews its release policy.
Read sourceShow 2 older updatesHide older updates
Tavus AI model Griffin scores 3.83 on NVIDIA face-to-face benchmark, near human level
A company named Tavus has built an AI model called Griffin that is capable of passing as a human in live conversation. According to the post, 48% of people who interacted with Griffin live believed it was a real human. The model scored 3.83 out of 5 on NVIDIA's scoreboard for face-to-face AI, closely approaching the human benchmark of 3.92. The next best competing system scored only 2.80. The post claims that unlike earlier systems that stitched separate models together for hearing, thinking, speaking, and animation—which introduced delays at each handoff—Griffin performs all these functions in a single system. It can read a user's face, tone, and words, and react while the user is still speaking. The author, who received early access, agrees with the assessment that the model is highly realistic.
Read sourceTavus study: 45% mistook AI for human in video call; 28% doubted real humans
A study by Tavus found that 120 people watched a 90-second video call and 45% believed the person on screen was real, but it was an AI. In a separate test, 120 people watched two real humans talk, and 28% thought one was a machine. The company gave early access to its smaller model, Griffin Lite. The author notes that looking human on video has become a timing problem, as humans leave roughly 200-millisecond gaps between turns, while turn-based voice AI often gives itself away by pausing. Tavus describes the model as full duplex, meaning it watches and listens while talking, can be interrupted, and notices people entering the frame. NVIDIA researchers created a benchmark from 237 real two-person video call clips, where real humans score 3.92 out of 5. Tavus reports its model scores 3.83, compared to 2.80 for the best published system. The model responds in about 1.9 seconds, versus 2.8 seconds for the other system, while humans answer in 0.9 seconds. The author concludes that language models learned what to say, and face-to-face AI is learning when to say it.