Cocho

सर्वैः नरैः सेवितव्यम्
El muerto al pozo y el vivo al gozo
Libertarian Nationalist🎖️

[Cocho @CochoFaucet]

AI is here to stay, but...
Cognitive bankruptcy + the fatal token subsidy cliff will hit faster than killer apps profitability or AGI.
There's overproduction of low-value work and underproduction of high-value, defensible business models.

[UHN Plus @UHN_Plus]

🇦🇷‼️ | Javier Milei oficializó una estrategia geopolítica y jurídica orientada a transformar al país en la capital mundial de la innovación tecnológica. A través de un manifiesto publicado en el Financial Times, el mandatario expuso la necesidad de romper con el modelo de imitación de los centros de poder tradicionales para anticipar el futuro, proponiendo una arquitectura legal revolucionaria que otorgue personería jurídica y responsabilidad limitada a entes no humanos gestionados por algoritmos.

[Mathematica @mathemetica]

Camillo De Lellis (born 11 June 1976) is an Italian mathematician and the IBM von Neumann Professor at the Institute for Advanced Study.

He received the 2026 Shaw Prize in Mathematical Sciences and is active in the fields of calculus of variations, hyperbolic systems of conservation laws, geometric measure theory and fluid dynamics.

His achievements include deep results in geometric measure theory on the regularity of minimal surfaces and currents, as well as analysis of singularities in fluid dynamics, including the Navier-Stokes equations.

De Lellis has resolved long-standing questions about the structure of singularities. He continues work on geometric analysis and PDEs.

[Sandro @pupposandro]

Excited to launch Luce KVFlash. We've been working harder than ever with @davideciffa to bring better DX for local AI.

Today, long context has a second memory bill nobody budgets for: the KV cache.

On Qwen3.6-27B at 256K it costs 4.6 GiB of VRAM and drags decode down to 13 tok/s, because every new token reads the whole thing.

KVFlash keeps a small pool of KV on the GPU, auto-sized to your VRAM, and pages cold 64-token chunks to host RAM, bit-exact and recallable.

decode holds a flat 38.6 tok/s from 64K to the native 256K on a 3090, 2.9x the full cache at 256K, 72 MiB resident and benchmark accuracy unchanged.

[Brian Roemmele @BrianRoemmele]

Imagine a future dominated by automation, where traditional jobs vanish, citizens invent fake job categories to avoid unemployment, exposing societal absurdity.

This urges embracing creativity and innovation rather than bureaucratic solutions.
https://t.co/Jn2Fi6olIJ

[Cocho @CochoFaucet]

@ivanfioravanti You need @Starlink

[Lotto @LottoLabs]

Everyone is gonna be localmaxxing soon 🤗

[Lotto @LottoLabs]

Keep it simple if you want to learn local ai

1) Build the cheapest rig that can house 1 used 3090

2) Who cares about ram, cpu, etc, buy ddr4 32gb, get a decent mid range ryzen/intel

3) download qwen 3.6 27b

You’ll be into the rig like $1500 at most

You won’t be able to upgrade much (maybe another 3090) but it’s a great starting point

You don’t feel locked into learning about big rigs and upgrade paths

If you like it then you can still keep and run the 3090 rig (trust me it’s useful)

And you can build a new separate rig if you need more vram, on any platform after you have some experience and know what type of hardware you actually need, not what people recommend online!

[Lotto @LottoLabs]

90% of pretraining is just getting nonslop datasets and curating by hand

You can train a 1B model over a couple weeks on a 3090

Once you have the data you can plug and play different architectures

Just start bro

(Also another reason I like the 3090 for inference, you can actually train with it too)

[Mashood tried Ops @fromcodetocloud]

Python + shell, AWS, Jenkins, Github Actions, Terraform, Docker, Kubernetes, Prometheus, Grafana

-> Devops Engineer

[Ahmad @TheAhmadOsman]

Kimi-K2.7-Code is the new Opensouece SoTA for Coding & Agentic workflows

[Mashood tried Ops @fromcodetocloud]

If you are not much interested in development and have a chance to choose between a career in Development and DevOps/SRE,

Choose DevOps/SRE

[Brian Roemmele @BrianRoemmele]

The morning of July 24, 1915, the SS Eastland rocked gently at the Chicago River dock between Clark and LaSalle.

Her decks were alive with 2,572 excited passengers, mostly Czech and Polish immigrants and their families, headed to a Western Electric company picnic.

Laughter filled the air. No one noticed how dangerously top-heavy the old steamer had become.
At 7:28 a.m., still moored, she lurched and rolled ninety degrees, capsizing into twenty feet of filthy water.

The joy turned to horror in a single deafening roar: 2,500 people screaming at once. Hundreds were trapped inside the hull; hundreds more thrashed in the river. Heavy skirts dragged women and children under like stones.

Fifteen-year-old Helen “Nellie” Nowak, the night elevator girl at the neighboring warehouse, froze on the dock.

She had been hauling produce since 4 a.m. and couldn’t swim a stroke. The sound hit her like a blow. Without hesitation, she raced inside, grabbed forty feet of thick Manila rope, tied it to a steel bollard, and flung the other end into the chaos.

For forty-five brutal minutes, Nellie lay on her stomach and pulled. She looped the rope around wrists, braced her feet, and hauled people up, sometimes three at a time.

Her palms tore open to the bone. When the pain became unbearable, she tied the rope around her own waist and kept going, teeth clenched, body screaming.
Police finally dragged her away at 8:30 a.m. By then, Nellie had saved twenty-one lives.

Her hands were ruined, a rib broken. Western Electric gave her a job for life; she worked there fifty years.

When companies had a soul.

The survivors called her “The Girl With The Rope.” That blood-stained Manila rope now rests in the Chicago History Museum, a quiet testament to one ordinary girl who refused to let go.

[Mandrilandia @MandrilandiaX]

Algunos acá se sobregiraron y se indignan que entre nosotros no puedan haber disidencias. A mi nadie me va a decir qué puedo opinar y qué no. Encima muchos de esos se la dan de “liberales intelectuales”. El chiste se cuenta sólo.

[Unsloth AI @UnslothAI]

Gemma 4 now runs 2x faster with MTP GGUFs! Run locally on just 6GB RAM. ⚡️

MTP enables Google Gemma 4 run ~1.4–2.2× faster with no accuracy loss.

Gemma 4 12B MTP can run at 162 t/s vs. 52 t/s without MTP. 31B reaches 101 t/s.

GGUFs + Guide: https://t.co/c4gAUlb6YE

[Unsloth AI @UnslothAI]

DiffusionGemma can now run at 2000+ tokens/sec! ⚡

We made local DiffusionGemma inference 1.8× faster.

Run it on 18GB RAM via Unsloth Studio.

GitHub: https://t.co/aZWYAtakBP
Guide: https://t.co/wYLfJWE6kG

[Unsloth AI @UnslothAI]

MiniMax M3 can now be run locally!🔥

MiniMax-M3 is a new 428B (23B active) open model with 1M context that performs on par with Gemini 3.1 Pro.

Run Dynamic 2-bit GGUF on 138GB RAM/VRAM or 3-bit on 165GB.

GGUF: https://t.co/lwfWsOBNKl
Guide: https://t.co/EP62nmKK0R

[jinjingliang @JinjingLiang]

@AnthropicAI The state of things:

[Ahmad @TheAhmadOsman]

https://t.co/riIctxjHoN

[Ahmad @TheAhmadOsman]

Equating opensource models to nukes and nuclear weapons was the greatest psyop of Anthropic

Technology has always been evolving on all sides, stop it with that crap that’s meant to control you

[Ahmad @TheAhmadOsman]

Fuck you, @AnthropicAI, you guys are evil and malicious and you know exactly what you're doing to secure your moat through regulatory capture