Yaz retweeted
today, we’re reintroducing @arcstudio.
i started Arc as a 16-year-old high school student who loved design.
since then, we’ve grown into a team of four, worked with some of the fastest-growing companies in the world, and generated $200k in revenue.
we’ve had the privilege of working with @automattic, @axiomexchange, @orchid_hq, and many others.
today marks a new chapter for Arc, with bigger ambitions and a lot more to build.
thank you to everyone who’s helped us get this far.
we’re going to the moon! 🚀
Yaz retweeted
I was able to get early access and damn this is so cool
Naturally, my mind navigates to how visual AI will play a role in the personal AI assistant space.
I think there are legitimate applications, such as:
1/ identify an object if I don't know what it is
2/ help me fix something, in real time, with visual feedback
3/ help me prep for an upcoming pitch
4/ give me feedback on my clothes, am I matching?
My video below shows how the AI performed with each prompt.
Helping me troubleshoot why my TV wasn't working was definitely the most impressive to me
Once visuals like this enter the game, we're going to see a whole new level of disruption, I can't wait
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).
Readers added context they thought people might want to know
The 48% figure and "video Turing test" claim are from Tavus's own study of 54 one-minute calls, not independently verified or using a standard protocol. Griffin-Lite leads NVIDIA's VideoFDB benchmark on their public leaderboard.
cellcog.ai/blog/tavus-gri…
research.nvidia.com/labs/amri/proj…
tech-ish.com/2026/10/02/tav…
Yaz retweeted
They list some examples in the replies: tutor, companion for the elderly, humanlike assistant. I don’t see anything unclear or wrong about making these things more broadly accessible
It even wrote the score generator and its rhythm/harmony checks in Py. Every export was then independently checked via Music21 and Mido against strict voice-leading, harmonic, and rhythmic rules delegated to a couple Sol subagents.
Once validated, DawDreamer took routing CC0 multi-samples, automating continuous expression curves, panning, and spatial reverb. Exported headlessly to 24-bit WAV with triangular dither.
Yaz retweeted
since @Hetzner_Online raised their prices and has low availability, here is a solid substitute:
netcup.com
Insanely cheap prices, solid performance, and traffic included
This is completely normal. The market decides what lives, and the power law is doing exactly what it's supposed to do. People are freaking upset for them because a Harker and Caltech grad chose to build this in a couple weeks last year instead of something noble. But these founders don't need to sit around unemployed faking noble passions just to stand out. They don't have to carry the heavy chips on their shoulders that drive the rest of us, whether we're mostly self-diagnosed megalomaniacal, delusional, or autistic kids of the popular culture. Just like prime 2010s SV, these kids stepped into the arena. They proved their drive to SF, took YC strictly on merit, and used their high self-esteem and raw agency to get exactly what they needed out of it. Instead of being upset for them, we should be envious. I know I am. Win or lose, they'll be absolute powerhouse hires for any consumer, entertainment, or devtool co. Idk any top 100 startup that wouldn't hire them on the spot, regardless of the role they get into.
This quoted post is unavailable.
here's what actually happens, always:
founder-market fit is a big deal. yc loves it, the valley loves it. but if users don’t actually want what you’ve made, none of that matters. credentials open doors, sometimes a few too many. i’ve observed some pedigreed teams build for themselves, not the market, and get claps from investors but not users.
saw quibi burn billions this way. also saw misfits launch weird things, get no respect, pivots, then the market floods in and suddenly it’s undeniable (see: scale ai, airbnb).
users either pull the product out of you or they don’t. if you have to brute force your way in with fundraising and background, that’s not signal, that’s noise. usage is the only scoreboard that counts.
if people want “brainrot slop” in their ide, the data will show it. if not, it’ll fade. keep shipping, keep measuring. credentials might get you in the batch; only product fit keeps you alive in the long run.
product > background, every single time.
