@iNoraF1 Que hayas tenido un lindo día 🌸 🙏 🫶
Brushy Creek, TX 🇺🇸 Español
Romeo Cabrera Arévalo
111.6K posts
Romeo Cabrera Arévalo
@romcabrera
Artificial Intelligence and XR research and engineering
Austin, TX Katılım Temmuz 2009
562 Takip Edilen1.9K Takipçiler
Mi mami se acordó que hoy es mi cumpleaños y es el mejor regalo que me han dado en mi vida.
Español
Baidu just open-sourced an OCR model that reads entire 40-page documents in one shot.
It's called Unlimited-OCR. 3 billion parameters but only 500 million active during inference. Runs 100% locally on your machine.
Why this matters: traditional OCR tools chop documents page by page. Tables that span two pages break. Reading order gets lost. Cross-page context disappears.
Unlimited-OCR processes the whole document at once. 32K context window. Text, formulas, tables, reading order all preserved across pages.
Output comes out as clean structured Markdown.
→ 93% accuracy on the standard benchmark. +6 points over the baseline.
→ Error rate stays below 0.11 even past 40 pages.
→ Multilingual out of the box.
→ 2.12 million downloads on Hugging Face last month. 14,600 GitHub
stars.
For context: Amazon Textract, Google Cloud Vision, and Azure Document Intelligence all charge per page. This runs locally for free.
English
Romeo Cabrera Arévalo retweetledi
In Texas they told me to stop at Buc-ee's for gas.
I have been to shrines. I have stood in temples that took two hundred years to build.
I was not prepared for the gas station.
There were one hundred and twenty fuel pumps.
I counted them because I did not believe them.
A man beside me was filling a truck the size of my first apartment, and he was not filling it because it was empty.
He was filling it because he was here, and here is where a man fills things.
Inside was a hall so vast I lost the horizon.
A wall of jerky. A wall of fudge I did not know the country produced.
A brisket sandwich handed to me by a man in a beaver costume.
And I want to be clear, the beaver is not a mascot.
The beaver is a saint.
The people speak of him the way my grandmother spoke of the mountain behind her house.
And the bathrooms.
I had been warned about the bathrooms and I had dismissed the warning as the pride of a loud people.
I was wrong to dismiss it.
The bathrooms are famous across the whole state and they have earned it.
I have slept in worse hotels. I nearly bowed upon entering.
A janitor was polishing the floor with the devotion of a man tending a garden he loved, and when I thanked him he said "welcome in," which I have since learned is what Texas says instead of hello, and also instead of I am glad you exist.
I went in for gas. I was inside for ninety minutes.
I came out with fudge, a shirt printed with a joke I do not fully understand, forty dollars of jerky, and a feeling I can only describe as having been to church.
I did not need any of it.
I needed all of it.
I have walked through the great cathedrals of the old world. I lit no candle there.
I lit no candle at Buc-ee's either.
But I did fill the truck.
And I understand now that in Texas, this is the same thing.
English
Romeo Cabrera Arévalo retweetledi
@kuskunga Marcar goles no es solo cuestión de que les digan "ha galo".
Brushy Creek, TX 🇺🇸 Español
@kuskunga Eso se debe en bona parte a las decisiones del cuerpo técnico.
Brushy Creek, TX 🇺🇸 Español
Brushy Creek, TX 🇺🇸 QME
𝗔𝗛𝗢𝗥𝗔 Erling Haaland llega a Noruega cargando un mapache. 🦝🇳🇴
Español
Romeo Cabrera Arévalo retweetledi
Jude Bellingham taking it all in as thousands sang “HEY JUDE”. A special moment. 🏴
English
Romeo Cabrera Arévalo retweetledi
Romeo Cabrera Arévalo retweetledi
BREAKING: $META is reversing the decision to allow Ai to be used on any public instagram profile
English
Romeo Cabrera Arévalo retweetledi
I've been wanting proper Gaussian Splatting support in @rerundotio for the longest time. I had previously vibe-coded a custom view that kind of worked before, but it had some wonky bugs and was still too rough around the edges. This was originally done back in the Codex 5.4 era, and as you can imagine alot has changed. So I decided to revisit the implementation because of the following:
- Rerun 0.34 release brought us MCP support for the rerun viewer. This means an agent can take screenshots, interact with the viewer, rotate 3d spatial components, and open/close views. Basically, anything you can do in the rerun viewer, an agent can now do, closing the loop on debugging.
- Significantly stronger models have been released from both Anthropic and OpenAI. Fabel and Codex 5.6 Sol should make this implementation both cleaner and more performant, as well as find the previous bugs 5.4 couldn't
- Harnesses have gotten much, much better since I first worked on this. HTML artifacts + /goal have made implementation + review much easier. I actively have Fabel in the driver's seat while 5.6 does mechanical code work + review after the fact.
Thanks to the combination above, I was able to improve the custom Rust view to be lots better and fix most of the wonky bugs. Along with this everything should now use proper GPU rendering with no CPU fallbacks. I also wanted to make sure I could visualize training so I went ahead and integrated it with the wonderful brush since it works on Mac as well.
With this proved out, we're also working on getting official Gaussian Splatting support as a proper archetype on the normal rerun distribution. This should be a great foundation to build all of the other esoteric representations (2DGS/4DGS/Triangle Splatting/whatever else you want)!
I can finally log the full splat training run and move forward and backwards in time to see how the optimization went at different timesteps!
Given powerful agents, the rerun mcp client, and a good harness, there's literally never been a better time to build.
Pablo Vela@pablovelagomez1
There's been a few cool updates recently. In particular, @rerundotio 0.33 released headless rendering. This, along with the Fable 5 release pushed me to work torwards making MAMMA realtime! I threw Fable at the problem, and it was able to take original implementation that was ~12 seconds / frame and get it all the way down to 40ms /frame, or nearly a 300x speedup 🏎️ How did I achieve this? TLDR: - Use rerun's headless rendering as supervision when optimizing - Save rrd file as test fixture to guide model optiziation with /goal - create an html artifact with headless rendering to provide detailed breakdown of what it did and how it actually looks like in the viewer There were a few critical bits to make sure that this ACTUALLY worked and that Fable didn't just cheat or delete something and declare victory. The first is that the original version used Rerun, this allowed us to save things to disk as an RRD file, meaning we could query the contents and use this as a sort of test fixture or golden artifact that held EXACTLY what all of the values should be. Then we can use this with /goal as a metric when doing the optimization to ensure there are no regressions. The second bit is the headless rendering, this gave us the ability to check that not only did the test fixture pass, but it also looked visually correct. This made a huge difference, and an awesome side affect of it is that we can use the headless rendering to create an implementations.html file. This gives a visual guide as to what the agent did (I walk through it in the video below) Along with this, we're working on an MCP server for rerun that allows full interactivity with the rerun viewer for your agent. So for example the agent can click, drag, move views, scroll timelines, ect. I used this to help the agent debug certain parts such as when the 2d sam masks didn't line up, or if the triangulated keypoints werent correctly matching with the optimized mesh. The agents could go, click into the view, scroll through the timeline and see where things went wrong. Fable + Headless Rendering + Rerun MCP == 300x speedup in less then a days work With these new tools, I'm planning on going back to my gaussian splatting implemntation and cleaning it up + making it fast!
English
@washingtonpost Why we have to take extra steps for something they arbitrary chose to do?
This is some violation of our privacy and @Meta is scot-free doing whatever they want
English
Meta has launched a new AI image generator that allows anyone to create AI images with your Instagram photos. And if you have a public Instagram account, Meta automatically opted you in.
Here’s how to opt out: wapo.st/4ymzOnN
English
@2010MisterChip Lukaku debe estar liderando algún tipo de estadística "cantidad de goles por cada 60 minutos jugados", se podría determinar? Tantos goles sin jugar partidos completos en este mundial...
Port Aransas, TX 🇺🇸 Español
Ahí tenéis “al muerto de Lukaku” haters. No para de callar bocas este pequeño gran hombre.
Español
@mariapazza De lo que entiendo es un concurso de popularidad, votan los fans, no es algo técnico play.fifa.com/potm/
Brushy Creek, TX 🇺🇸 Español
El penal… bueno, debatible.
El offside: ROBO ROTUNDO de Portugal.
Pero que le den el MVP a un jugador que no dio ni UN toque efectivo hasta que le dieron el penal: VERGONZOSO.
Español
@antoni_doroteo @2010MisterChip Que un equipo tenga a los mejores arqueros de su país... para que uno sea suplente. Lo hace menos probable
Brushy Creek, TX 🇺🇸 Español
@2010MisterChip ¿Y con la misma selección? Es mucho más difícil el dato porque aunque sea lógico que los jugadores de una selección puedan jugar en los equipos de su liga nacional, al mismo tiempo es casi imposible que lleven dos arqueros del mismo equipo, y que además se reemplace el portero.
Español
Solo 3 veces en TODA la historia de la Copa del Mundo dos porteros del mismo club han jugado el mismo día con distintas selecciones:
22.06.2006 - AC Milan: Dida 🇧🇷 + Kalac 🇦🇺
23.06.2010 - Wigan Athletic: Kingston 🇬🇭 + Stojkovic 🇷🇸
01.07.2026 - LE HAVRE FC: Mory Diaw 🇸🇳 + Mpasi 🇨🇩 ← HOY
Español
Estaba cantado que iba con el mismo 11 que contra Alemania.
Español
¿Cómo que el gentilicio de Países Bajos no es "pordebajeados"?
Wells Branch, TX 🇺🇸 Español
This World Cup has given me a masterclass in intra-Latin racism so intense it’s giving intra-Asian racism a genuine run for its money.
Truly elite work, folks.
Pedro@PedroSMR650
Los Argentinos llegando a defender a los Ecuatorianos de los insultos racistas de los Mexicanos, con más racismo.
English