People are beginning to understand why the heads of the frontier labs don’t eat animals.

Yesterday we previewed Griffin, a Human Interaction Model capable of seeing, hearing, sounding, and looking like a human does. There have been a lot of questions, so I wanted to take a moment to share our thoughts. By way of introduction: Tavus is a research lab focused on enabling machines to meet us where we are, and to understand the nuances of how we communicate beyond words. Griffin, our latest model, isn’t publicly available yet. Yesterday’s announcement was a limited research preview to demonstrate what the model is capable of. With AI progressing so quickly, we prefer to share breakthroughs openly and in real-time as we work on a safe public release. Face-to-face is how we evolutionarily communicate, it carries the most meaning and intent, and we want computers to be able to help with work that benefits from that emotional understanding, expression, and immersion. Some examples of the kinds of use cases we care deeply about: - A tutor that can build understanding of how a student learns, see exactly when there is confusion or disengagement, and adapt the lesson to fit them. - A health expert that can answer any questions about your upcoming appointment or prescription, at the pace you want, at any time you need, even on a weekend. - A language coach that you can practice speaking with, that can correct your movement and pronunciation, and help build confidence to have real conversations - Or the perfect assistant for everyone, that understands intent, knows how you work and remembers what matters. We’re working with partners on safeguards and systems for disclosure, as well as inviting discussions with officials around wider regulation and safe use. People will always know they’re interacting with AI, while providing an interface that removes the need to ‘speak computer’. We believe in a future where computers understand us well enough to make technology more accessible, more useful, and make us more capable as humans. That is the world we want to build, and we understand the responsibility to do so safely.

One day I hope to go so viral that Bernie Sanders tweets out against our company existing. Truly incredible

Think of how every tech is scary at first but then we figure out the guardrails (scams are hard to do at scale with the costs of running this model, and that'll be true for a while) and we figure out norms (we expect people to *know* it's ai with disclosures). We don't ban email because it's used for phishing or cat fishing. We have safety mechanisms in place that we build with better engineering. The benefit of this tech is that the machine tells that distract you and lead to an uncanny valley where you have robot voice and don't think or act naturally fall away and you can engage with your ai math tutor, or your ai role playing dungeon master, or your sdr/bdr trainer, or your elderly companion for those periods between visits. Like all new tech, this is scary until it's not because we engineer the safety that makes it fit in with society's values.

Real-time video is solved

Introducing Griffin, the first model to pass the video Turing test. 48% of people who talked to it live thought it was a real human. Previous systems have had a pass rate <3%. It is #1 on NVIDIA's benchmark for full-duplex AI video. It’s the first Human Interaction Model (HIM).

Benchmarks put Tavus squarely at #1, and that’s with the released models you have access to.

I’ve been sporting an iPhone SE for the past three generations and I’m finally upgrading to the Duo. Not at all put off by the Touch ID since that’s what I’m used to… happy about it in fact. Seen enough goofy faces trying to unlock phones to know that I prefer a Touch ID that doesn’t need my phone or face contorted.

The “empty calories” framing is an insightful way to think about the increasing number of PRs I’m layering on top of my repos. Statistically nobody’s batting 1000, so if you’re not rejecting PRs you’re introducing weakness and incoherence. Rejecting PRs is a necessary step to honing taste and understanding of how your repos should be operating.

Glad we submitted, #1 on TurnBench @code_brian pulled it off

I'm interested in moving my many, many parallel agents to the cloud and want a way to handle my half dozen mcp servers in a nice way so I can auth locally and have it work everywhere. What's your preferred workflow for cloud agents? I use a mix of Claude and Codex and now Grok. I also want to have agents cut, crop, and post screenshots, so they need a browser. Locally, I have computer use handle it. But it doesn't scale and I'm tired of bogging my computer down once I hit 6+ concurrent agents. It's so clear to me that cloud agents that work with both browsers and mcp servers will unlock dozens of concurrent agents, because at this point I have a huge backlog of linear tickets and plans and I want PRs with screenshots out the other end.

I love not having to think about how a robot hears differently than a human

1M views 🥰

This is what vibe coding looks like for human-like AI agents. Tavus just launched PAL Maker. Instead of manually configuring the face, voice, personality, knowledge, memory and deployment of a video agent, you simply tell Charlie what you want to build, and he creates it with you. A few clicks later, your PAL can be shared or embedded into a website or app. This is the important shift: Tavus is turning real-time, face-to-face AI agents from a developer project into something almost anyone can create. I think this could unlock an entirely new wave of AI applications. This is literally how I imagined the future to be.

this is super freaking cool. ngl, this is how i imagined the future!