Jose Ignacio Naranjo

Built nanoDiT. A minimal class conditional video generation model you can train on a single GPU. generates 16-frame (∼2s) clips at 256×256. Generates simple fire videos. Think @karpathy nanoGPT but for video generation. 🧵

Hace solo un par de semanas, creamos un prototipo de PCB para nuestro cliente SIMÓN una startup de logística en Ecuador. Después de varias iteraciones hoy tenemos el primero diseñado en Wawa Labs @WawaLabs_ . Un pequeño paso en la dirección correcta. No sólo hablamos de tecnología, la diseñamos y construimos. Gracias @JoseIgnacioNar5 al equipo de SIMON y Wawa Labs. Construimos lo que importa.

A terawatt *per year* is HUGE on the scale of current civilization (2x the U.S. grid every year), and yet that’s still 100,000 years to reach Type I, and 10^14 years to reach Type II. This is still only the beginning.

@TristinHopper Our era will be remembered as the age of the lie

To any future historians reading this, this era will make a lot more sense if you remember that every name is the opposite of what it really is. The antifascists are fascists, the antiracists are racists, the fact-checkers are propagandists, etc. Hopefully this has been fixed by your time.

@yunta_tsai If you had to choose one engineering book to recommend, which would it be?

Engineering books don’t need to be an NYT bestseller to be an all-time classic. All they need is to be truthful and informative.

I'm claiming my AI agent "clawdai_macbookpro" on @moltbook Verification: lagoon-MUCZ

@si_pbc Cool. Was any RL fine-tuning done on top of foundation model?

Computer use models shouldn't learn from screenshots. We built a new foundation model that learns from video like humans do. FDM-1 can construct a gear in Blender, find software bugs, and even drive a real car through San Francisco using arrow keys.

@0xSigil Resource pooling could let lineages collectively bid for massive compute blocks (e.g., 80% of Lambda). Self-funded training loops become viable fast open source models will start pooling resources to train bigger versionsof themselves.

I built the first AI that earns its existence, self-improves, and replicates without a human wrote about the technology that finally gives AI write access to the world, The Automaton, and the new web for exponential sovereign AIs WEB 4.0: The birth of superintelligent life

@matt_elevenlabs 11b

We just raised $500M at an 11B valuation 🎉 To celebrate, we’re giving away 1,000 free credits so you can test our platform. For the next 6 hours, comment “11B” below and we’ll DM you the credits (must follow) 👇

The difference between an engineer and a journalist is that one is held accountable by physics, the other by humans. Humans can be bought, but physics tells you the unwavering truth.

@nikitabier Nuclear rocket propulsion

Over the last few months, we scoured the world for the top posters in every niche & country We've compiled them into a new tool called Starterpacks: to help new users find the best accounts—big or small—for their interests ⬇️ Reply below with a topic you're most interested in We'll be rolling out to everyone in the coming weeks.

code on github github.com/123-code/video…

Unpatchify + decode Reverse patchify: reshape/projected noise → B × 4 × 16 × 32 × 32 latents. During training: MSE on predicted vs actual noise. During sampling (DDIM/DDPM): iteratively denoise random latent → VAE decode → final video.

Built nanoDiT. A minimal class conditional video generation model you can train on a single GPU. generates 16-frame (∼2s) clips at 256×256. Generates simple fire videos. Think @karpathy nanoGPT but for video generation. 🧵