spanish designer in Paraguay | @Duelbits [email protected]
Joined April 2017
- Tweets2.3K
- Following898
- Followers14K
- Likes4.9K
clay. retweeted
wait, sunra's enterprise deals go up to 40% off leading models' own API prices
not off some reseller markup. off what the labs themselves charge
clay. retweeted
before: a subscription, an account with every lab and a bill from each
now: one sunra API key, every model at its maker's own price, pay as you go
cheapest way to cool a data centre in 15 seconds for about 40 cents
the interesting part is that it followed every rule it was given
Introducing Bivia.
A multimodal AI decision model built to complete entire workflows, not just answer individual questions.
We've spent 3+ years building and deploying AI in production across semiconductors, energy, manufacturing, and other complex industries.
Today, we're bringing that experience to everyone.
clay. retweeted
Introducing Bivia.
A multimodal AI decision model built to complete entire workflows, not just answer individual questions.
We've spent 3+ years building and deploying AI in production across semiconductors, energy, manufacturing, and other complex industries.
Today, we're bringing that experience to everyone.
Jensen Huang, CEO de NVIDIA acaba de explicar por qué un agente de IA necesita algo más que un buen modelo para hacer el trabajo de una empresa.
Pasó más de una hora en directo sobre el escenario explicando cómo realizar (en la mitad de tiempo) tareas que a la mayoría de los equipos les llevan semanas.
Esto vale más que cualquier curso sobre sistemas basados en agentes, por 997 dólares que cueste.
Quienes aprenden a crear agentes están desarrollando por su cuenta aquello que los demás delegan a un equipo externo.
introducing prism, the world’s first speech-to-speech model you can’t tell from a human in conversation
prism has:
> 300ms response time, 4x faster than OpenAI GPT-Live
> outperforms every other s2s model on intelligence (with NO reasoning)
> the first s2s model customizable through reinforcement learning
we’re a team of 35 with $35M in funding, and we’re unlocking s2s model adoption faster than openai, google, and x.
the first 100 people to sign up here get 1 month free: rime.ai/prism?utm_campaign=p…
clay. retweeted
introducing prism, the world’s first speech-to-speech model you can’t tell from a human in conversation
prism has:
> 300ms response time, 4x faster than OpenAI GPT-Live
> outperforms every other s2s model on intelligence (with NO reasoning)
> the first s2s model customizable through reinforcement learning
we’re a team of 35 with $35M in funding, and we’re unlocking s2s model adoption faster than openai, google, and x.
the first 100 people to sign up here get 1 month free: rime.ai/prism?utm_campaign=p…
clay. retweeted
this means you can basically clone any AI platform
and use all these AI models for half the price..
filming a real time video model live on set is a bold call. one bad frame and the whole take is gone
curious how many takes didn't make it into this cut
clay. retweeted
ESTO ES LO MÁS ADICTIVO QUE HE VISTO EN TODA LA SEMANA
Una oficina entera en miniatura.
Gente llegando, sentándose, abriendo Outlook, saltando a Teams, editando documentos… a cámara rápida.
Y no es una animación cualquiera.
Es HelixCove, un dataset que acaba de publicar Arbium para enseñar a la IA cómo se trabaja de verdad.
Lo llaman "human traces": el rastro de todo lo que haces en el ordenador.
Qué abres, en qué orden, a quién escribes, cómo terminas una tarea.
Y aquí viene lo fuerte:
→ Le hicieron más de 400 preguntas sobre la empresa a los mejores modelos de IA
→ Sin documentación, sin conectores, sin explicarles nada
→ Solo mirando cómo trabajaba la gente
→ Y aun así acertaron la mayoría
Imagina una IA que entiende tu empresa como un compañero que lleva años allí.
Sin que tengas que contarle el contexto cada mañana.
Todavía es una preview.
Pero da una idea muy clara de hacia dónde va esto.
before: a subscription, an account with every lab and a bill from each
now: one sunra API key, every model at its maker's own price, pay as you go
clay. retweeted
I think about this every single day. Lee Kuan Yew took over a tiny island with no natural resources and people who didn't even speak the same language, and turned it into one of the richest, least corrupt countries in the world. All within 30 years. It's absurd this isn't mandatory study for anyone taking public office.
What's so special about Singapore's success is how much common sense there was in LKY's approach. After taking over, he witnessed how having different cultures, each with their own language, was leading to a dysfunctional society. He not only fixed it by having all schools switch to a bilingual system, but he also chose English as one of the two. English was neutral to all cultures and the mother tongue of the future world leader.
There's this great passage in From Third World to First where he speaks about how he was "never a prisoner of any theory." The only test is reality. If you try something that doesn't work, you stop doing it and you double down on the ones that do. While visiting other countries, he'd recognize what elements of their foundations were good, and which were bad. The bar was whether it solved the issue for which it existed or not. Again, grounded in reality. Those factors that worked, he replicated in Singapore. And those that failed in other countries, LKY avoided.
It didn't matter to him if the USA or any other country did something. When creating the blueprint for Singapore's economic system, he didn't start with a traditional central bank, like every country had. LKY realized that unregulated issuance intrinsically bred trouble, so Singapore started out with a currency board. For each Singapore dollar issued, there had to be equivalent backing with foreign reserves.
The ultimate example of first-principles thinking was his reasoning for turning Singapore into a global financial powerhouse. Lee Kuan Yew observed that global trading started in Zurich, then Frankfurt and London, passed to New York, then to San Francisco, and after San Francisco closed nothing happened until Zurich reopened. There was a 6-8-hour gap, which Singapore could fill. By the '90s, its FX market moved hundreds of billions.
When Singapore had traffic bottlenecks, LKY started charging drivers to enter the city centre at rush hour. With that simple measure, traffic issues eased immensely. Similarly, to make Singapore a place people would be happy to visit, it had to be clean. Great effort went into cleaning rivers and planting trees, and measures like fining people for spitting in the street were introduced.
Over and over again, Lee Kuan Yew found solutions to extremely complex issues. Maybe one of the lessons is in what made this possible. Singapore's size allowed him to essentially be everywhere, because everything was close by. Maybe starting out from zero was also key. Trying this in a large country with all institutions in place might not lead to the same result.
clay. retweeted
Thank you.
To everyone who tried Sunra MCP, shared it, pushed it, and told us what broke and what made you smile: since launch, we've received a great deal of support from people like you, and it means a lot to our team.
And today, we're launching the Sunra API: every model at its maker's own price, pay as you go. This post is a bit on why we're building Sunra, and why we price it the way we do.
Before Sunra, we were on teams training our own models, back when nobody knew whether any of it would work. We wanted a new form of creativity and expression, and we pursued it with too few GPUs and too little of everything.
In the end, it was a numbers game.
But that's not disheartening. Models like Seedance proved that what we were all reaching for in the dark was real, and now more people can take up this new form of expression.
And it's still early. Model makers often can't see the full potential of what they've made. It takes creators, thousands and then millions of them, experimenting, failing and sharing, to find out. That's why education matters, and some teams already do this really well. Every good tutorial, every shared prompt and every strange experiment makes the next person's work a little better, and moves all of our imaginations a little further out.
None of that should be gated by cost, or by pricing designed to confuse. When every frame of film cost money, people thought twice before pressing the shutter. The price of a try decides what we dare to make. You should pay what's fair. Then you create more, more people learn, the whole industry grows, and everyone benefits.
Here's how our numbers work. We can count on future demand volume because of your purchases. Volume we can count on gives us leverage with the model makers, and we use it to negotiate prices below their own. We pass most of those savings on to you, because the cheaper a try, the more you make. The API gives you every model at its maker's own price, pay as you go.
Subscriptions on the Sunra app go a step further. We keep only what we need to keep doing what we love. It's still a numbers game, but this time the numbers are on your side. We've done the math, and we hope that the next time you press generate, you can flinch a little less.
When you don't have to think about the price, you get to think about the work, and that's what matters. Then we find each other, teach each other, talk late into the night about what comes next, and build it together.
It's still dawn for generative AI. We're glad you're here for the sunrise.
clay. retweeted
ok as a brazilian this one is actually useful
we don’t text, we send 3 minute whatsapp audios lol
a 2B model from Liquid that transcribes and translates them in 18 languages, running on my laptop, nothing leaving my machine?
trying it on my backlog tonight
🤝 Paid partnership
Introducing 𝙻𝙵𝙼𝟹: our new generation of Liquid Foundation Models. With a new tri-hybrid architecture and new multimodal encoders, the LFM3 family is the ideal model suite for specialized intelligence.
> 6 model instances, across text, audio, vision, and omni modalities, generative and decision models.
> 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱: text, 256k context window.
> 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱: vision-language model with video-input support for spatial understanding
> 𝙻𝙵𝙼𝟹-𝙰𝚞𝚍𝚒𝚘-𝟸𝙱: tool calls, voice chat, reasoning in 18 languages.
> 𝙻𝙵𝙼𝟹-𝙾𝚖𝚗𝚒-𝟸𝙱: an omni model combines vision, audio, and language.
> 𝚍𝟷-𝟸𝙱 and 𝚍𝟷-𝙾𝚖𝚗𝚒-𝟸𝙱: decision models with typed outputs and calibrated probabilities in a single pass. Omni adds audio input.
> fine-tune on 1 GPU.
> a new tri-hybrid architecture
> architecture optimized for 14 processor classes across GPUs, CPUs, and NPUs.
> 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱 is a beast. It performs competitively with Gemma 4 12B (6x larger), and Gemini 3.5 Flash Lite, and GPT-5.6 Luna, with 1.9B parameters.
> 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱 is #1 on the Pareto frontier of device intelligence 𝚙𝚒𝚙𝚎𝚝𝚝𝚎.𝚕𝚒𝚚𝚞𝚒𝚍.𝚊𝚒
Every modality for specialized intelligence. Pick the one that fits your use case, customize it, and run it on-device.
1/ 🧵