Roberto Núñez

Gerente de proyectos

[DW Español @dw_espanol]

La industria europea exige oxígeno: ¿modificará la UE el mercado de emisiones?

La industria alemana y europea enfrenta una tormenta perfecta: energía cara, competencia de China y Estados Unidos y una creciente presión regulatoria. Ante el riesgo de perder competitividad, Bruselas estudia cambios en el sistema de comercio de emisiones de CO₂.

[dunik @dunik_7]

a Tsinghua University lab just put a project on GitHub that replaces a $400,000 H100 rack with a single 24GB graphics card.

it's called ktransformers, and the trick is almost stupidly simple: the experts you actually use stay on the gpu, the ones you don't sit on the cpu until they're called

/ deepseek-v3 and r1 with 139K context in 24gb of vram
/ up to 28x speedup over the standard setup
/ fine-tune deepseek-v3 across four rtx 4090s instead of a datacenter
/ built by tsinghua university's madsys lab, not a startup with a landing page

apache 2.0, and already past 17,000 stars.

-> https://t.co/DjF8CutVAa

bookmark it.

[roman @Nozelcode]

🚨NVIDIA está regalando el uso de 5 modelos de IA china de última generación

Sin tarjeta de crédito
Sin suscripciones
Solo una API key que lo desbloquea todo

Lo que obtienes por $0:

→ DeepSeek V4: Flash para inferencia ultrarrápida
→ MiniMax M3: como asistente de código plug and play
→ Qwen3.5-397B: para tareas de razonamiento avanzado
→ Kimi K2.6: para flujos agénticos y cadenas largas
→ GLM 5.1: como modelo fiable para el día a día

Sin pagar suscripciones separadas por cada modelo

Sin cambiar tus workflows ni tus herramientas actuales

Sin vendor lock-in porque todo es compatible con OpenAI

Empezar te lleva menos de 2 minutos:

1. Entra en https://t.co/WZz4xNDezt
2. Regístrate y verifica tu cuenta
3. Genera tu nvapi key
4. Configura tu base URL en https://t.co/vIrWNmrMOp
5. Elige cualquier modelo y empieza a construir

Modelos soportados:

• Minimaxai: Minimax-m3
• Qwen: Qwen3.5-397b-a17b
• Moonshotai: Kimi-k2.6
• Zhipuai: Glm-5.1
• Deepseek: Deepseek-v4-flash

Pro tip:

→ Usa DeepSeek V4 Flash para velocidad.
→ Qwen para razonamiento duro.
→ Kimi para agentes.
→ MiniMax como tu compañero diario de código.

Lo mejor de todo:

Una sola key gratis te da acceso a más de 100 modelos del catálogo de NVIDIA.

~40 requests por minuto es más que suficiente para la mayoría de developers y proyectos personales

5 modelos frontera que compiten con GPT y Claude, completamente gratis.

Guárdalo y consigue tu API key gratis antes de que cambien los límites 👀

[🇨🇳XuZhenqing徐祯卿 @XueJia24682]

✨🇨🇳Xi Jinping:Where is humanity headed? All nations share the same boat. Instead of drifting apart on more than 190 small craft, we should unite as one big vessel or form a joint fleet, so that all mankind can sail toward a brighter future.

If human civilizations are reduced to a single color or a single model, the world would become a stereotype and too dull a place to live in. What we need is to respect each other as equals and say no to hubris and prejudice. We need to deepen understanding of the differences between our own civilizations and others, and advance exchanges, dialogue and harmonious coexistence among all civilizations.

[Superman @thesupermanmx]

China open-sourced a model that reconstructs any scene in 3D from a regular video, in real-time.

one camera. no LiDAR. 10,000+ frames without falling apart.

just walk around with your camera and watch the entire world get rebuilt in 3D at 20 fps.

→ runs at ~20 FPS on a single GPU
→ Stable over 10,000+ frames
→ Beats optimization-based methods on benchmarks
→ Works on drone footage, driving videos, indoor walkthroughs

100% open source.

[Li Zexin 李泽欣 @XH_Lee23]

Shenzhen just held the world’s first humanoid robot MMA event. This is absolutely the most futuristic city.

The white one even kicked the black’s head clean off in a spinning high kick! Brutal!

[Jota @jota_snchez]

Una profesora de Stanford lleva 20 años estudiando por qué unos tienen suerte y la mayoría no.

Su conclusión: la suerte es como el viento pero nadie sabe ponerle vela.

Si quisiera más suerte, haría estas 7 cosas:

1/ Micro-riesgos fuera de mi zona de confort.

[Dragon kingdom.wu🌃 @loongkingdom]

The West's attacks on Chinese manufacturing stem purely from the fact that Western-made goods cannot compete with Chinese products in any aspect—be it quality or technological application.

[liam. @liambraus]

el fundador de una empresa china de IA valorada en más de $20,000,000,000 acaba de dar una clase de 40 minutos sobre enjambres de agentes

la explicación más clara que he visto sobre sistemas de IA a gran escala

cámbiala por tus 2 horas de Netflix de esta noche

[Carlos Santana @DotCSV]

🔴 ¡NUEVO KIMI K3 OPEN SOURCE!

Primer modelo open source en ponerse por encima de lo que eran los modelos frontera de hace unas semanas!

K3 supera a GPT 5.5 y a Opus 4.8 quedándose no tan lejos de la nueva generación de modelos 🔥👀

https://t.co/wK55kKQtjM

[Mo Elgaraihy @EngMoElgaraihy]

الصين تضرب من جديد وتنهي عصر الهدر الرقمي! 🤯🇨🇳
​أطلق تحالف inclusionAI نموذجهم الثوري الأحدث Ling-2.6-1T، وهو وحش برمجي بـ ترليون بارامتر ، ليضع حدًا لجنون استهلاك الـ Tokens والأسعار الفلكية في النماذج الأمريكية!

​

[fred @heyfredds]

LE PIDES A TU IA QUE ENTIENDA TU CÓDIGO Y SE PASA 10 MINUTOS LEYENDO ARCHIVOS UNO A UNO.

Alguien construyó el atajo. Ya va por 86.000 estrellas en GitHub.

Se llama Graphify. Convierte tu proyecto en un mapa que tu IA consulta en vez de leer archivo por archivo.

→ Mapea código, PDFs, imágenes y vídeos con un comando
→ Marca cada conexión como real o inferida
→ Funciona en Claude Code, Cursor y 20+ agentes
→ Todo corre en tu máquina

Antes tu IA releía medio proyecto en cada pregunta.

Ahora traza el camino exacto entre dos piezas de tu código.

¿Tu IA se pierde en proyectos grandes, o ya tienes forma de ayudarla a orientarse? Repo abajo.

[Li Zexin 李泽欣 @XH_Lee23]

Europe urgently needs Chinese ACs, then no "overcapacity" now. Since June, they ordered 160,000 Midea PortaSplit.

On July 3, France added an emergency order of 20,000 units. Midea factory doubled its daily output from 3,000 to 6,000.

From order to 1st shipment: just 7 days.

[Superman @thesupermanmx]

NVIDIA open-sourced a 600M model that transcribes 40 languages in real-time at 80ms latency and it costs $0.

that's faster than you can blink. across mandarin, arabic, hindi, portuguese, tagalog, whatever,from a SINGLE checkpoint.

→ 17x more concurrent streams than buffered ASR on the same H100.
→ punctuation + capitalization built-in. no post-processing.
→ runs on your own GPU. no API bill

100% Open Source.

[Kshitij Mishra | AI & Tech @DAIEvolutionHub]

**Jeff Bezos just described AI in three words that make most of the economy temporary.**

Bezos said: *“AI is real and it is going to change every industry… it’s a horizontal enabling layer.”*

Three words.

**Horizontal enabling layer.**

That framing changes everything.

Most people still think AI is another app to use, another startup to invest in, or another productivity tool.

Bezos is pointing at something much bigger.

A technology layer doesn't sit beside the economy.

It becomes part of the foundation the economy runs on.

People keep debating which AI company will dominate.

That's missing the larger shift.

When a foundational layer changes, the companies on top either adapt to it or slowly become obsolete.

We've already watched this happen before.

First with electricity.

Then with the internet.

Neither immediately replaced existing businesses.

At first, everything looked familiar.

The same jobs existed.

The same companies made money.

The same rules seemed to apply.

Then, year after year, the entire system reorganized around the new foundation.

The winners weren't simply better competitors.

They were built for the new environment.

Everyone else was solving yesterday's problems.

AI feels like it's following the same path.

Not because it's another platform.

Because intelligence itself is becoming available like infrastructure.

Tasks that once required years of expertise can now be completed in minutes with the right AI systems.

Knowledge isn't disappearing.

The cost of accessing it is.

That's a completely different kind of transformation.

Entire industries have been built on one assumption:

Expert-level thinking is expensive, slow, and difficult to obtain.

As that assumption weakens, business models built around information scarcity will need to evolve.

The opportunity isn't just learning AI tools.

It's redesigning how you create value when intelligence is no longer the bottleneck.

Bezos wasn't simply making a forecast.

He was describing the direction the foundation is moving.

[Yarchi @undefinedKi]

Karpathy just did something that shouldn't be possible

He squeezed an entire GPT model into 200 lines of code. The whole working thing that trains and generates, with zero libraries and no GPU. Pure Python that runs on a laptop in about a minute.

This is the raw algorithm every LLM runs on, laid out on a single screen: the tokenizer, the attention, the training loop, the whole engine in plain sight

This is the clearest way to actually learn how an LLM works, end to end. Not a metaphor, the real mechanism. You finally understand what's happening instead of guessing, and from there you can build your own, on your own data, knowing exactly what every part is doing.

Free. Code, a web version, and a Colab you can run in your browser right now.

Repo: /karpathy/karpathy.github.io/blob/master/_posts/2026-02-12-microgpt.markdown

Bookmark this.

[CyrilXBT @cyrilXBT]

you won't F**king believe what this Anthropic engineer said on stage

"You're not supposed to prompt Claude. You're supposed to build a system that prompts itself."

In 45 minutes she breaks down how Anthropic builds agents that remember, learn from their mistakes, and get smarter with every run.

Worth more than any paid course you'll find on building agents.

Watch the session, then read the guide on building loops below.

[𝘊𝘰𝘳𝘳𝘪𝘯𝘦 @OopsGuess]

Forbes basically said the quiet part out loud.

For eighteen months, Washington called it “theft” when Chinese AI labs learned from American models.

Now an American lab founded by OpenAI’s former CTO releases a model that follows DeepSeek-style architecture and uses synthetic data from Chinese open-weight models like Kimi.

Suddenly it is not theft.
It is engineering.

Perfect.

Chinese open-weight models publish weights, lower costs, expand access, and let the world build on them.

American closed labs hide everything, charge premium prices, lobby for restrictions, and then accuse everyone else of stealing.

The irony is almost too beautiful.

The U.S. wanted AI to remain a closed imperial toll booth.

China turned it into infrastructure.

Now even OpenAI alumni are building on Chinese foundations.

That is not just competition.
It is gravity shifting.

[Deedy @deedydas]

Thinking Machines just dropped the best open weight AI model outside of China!

Inkling beats Nemotron 3 Ultra and benchmarks put it between Kimi 2.5 & 2.6. Many were contending to this throne, but Thinky has come out on top.

Really solid release, and will pair well with Tinker.

[Superman @thesupermanmx]

China just changed the game 🤯

They just dropped an open-sourced model that burns only 1% of the tokens compared to your favorite US models.

It's called Ling-2.6-1T and it goes toe-to-toe with Claude and GPT on agentic coding.

→ 1 trillion params
→ 72.2% on SWE-bench Verified
→ 256K context window
→ Free on OpenRouter
→ Plugs directly into Claude Code

100% Open Source.