Alvaro Enrique 🦦

Construyo cosas en internet

[Zaid @Ctrl_Alt_Zaid]

https://t.co/lKipGCmAR4

[Suraj Sharma @suraj_sharma14]

As an AI Infrastructure Engineer, you must build these projects.

Systems that prove you can serve, scale & keep GPUs honest.

1.) Self-Hosted Inference Cluster
Build: Multi-node GPU cluster serving an open model with vLLM or SGLang, health checks, and a public latency report.
Why: Talk is cheap. A live cluster is the portfolio.

2.) Cost-Per-Token Dashboard
Build: TTFT, ITL, GPU util (DCGM) and $ per 1M tokens by model, tenant and route.
Why: Infra without FinOps is just expensive uptime.

3.) Queue-Based GPU Autoscaler
Build: KEDA (or equivalent) scaling on queue depth, with cold-start mitigation and spot fallback.
Why: Idle GPUs kill startups. Slow scale kills users.

4.) Continuous Batching Load Test
Build: Traffic generator that stresses continuous batching, KV cache limits and saturation points.
Why: You don’t know your serving stack until it breaks under load.

5.) Model Weight Delivery System
Build: Safetensors registry + sharded weights + CDN/lazy loading for fast node bring-up.
Why: Cold starts are often a storage problem pretending to be a compute problem.

6.) Multi-Model AI Gateway
Build: Routing, retries, fallbacks, rate limits and per-tenant token budgets across 2–3 serving backends.
Why: Production isn’t one endpoint. It’s a traffic policy.

7.) Quantized Serving Bakeoff
Build: Same model in FP16 vs AWQ/FP8; publish quality vs latency vs VRAM tradeoffs.
Why: Optimization without benchmarks is cosplay.

8.) Secure Multi-Tenant Inference Layer
Build: Tenant isolation, keyed access, sandboxed tool/exec paths, audit logs.
Why: One noisy neighbor (or leak) ends the B2B deal.

9.) Checkpointed Distributed Training Job
Build: Ray or Spój/Slurm job with FSDP/tensor parallel, fault injection, and resume-from-checkpoint.
Why: Training infra is judged on recovery not the happy path.

10.) Speculative Decoding Prototype
Build: Draft + target model path with prefix caching / chunked prefill; measure accepted-token rate.
Why: Speed wins are product features when you own the stack.

11.) GPU Partitioning Lab
Build: MIG (or equivalent) slices with fair scheduling, resource limits, and contention tests.
Why: Real clusters share silicon. You must prove fairness under pressure.

12.) Triton / Custom Kernel Path
Build: One hot path accelerated (attention, sampling or tokenizer-adjacent) with before/after metrics.
Why: Senior infra energy = knowing when frameworks aren’t enough.

13.) Observability Spine for Non-Determinism
Build: Traces for queue time, prefill, decode, cache hits, errors alerts on drift and cost spikes.
Why: You can’t page what you can’t see.

14.) Multi-Region Failover Drill
Build: Active-passive or active-active inference with DNS/failover and RTO/RPO notes.
Why: “We have GPUs” isn’t reliability. Failover is.

15.) Public Benchmark Teardown
Build: Publish an architecture + latency/cost report others can reproduce.
Why: Hiring managers trust public artifacts more than résumés.

Most people watch tutorials. Builders ship systems.

Bookmark & Repost.

[Dario Amodei @DarioAmodei]

We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.

Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training.

You can read the full post here: https://t.co/OGyPb7yaYt

[ThePrimeagen @ThePrimeagen]

It is crazy how a technology that has the potential to cure cancer, alzheimer's, and who knows what else is so universally hated outside of rich tech.

How did Sama and Dario so thoroughly fumble this?

How did they not see telling everybody they're going to lose their jobs every few months and now we might (👉👈) accidentally kill everyone isn't a winning strategy?

[Maximiliano Firtman @maxifirtman]

🔴IA: MENTE ALIENÍGENA

El jefe científico de OpenAI publica un inquietante artículo sobre como ven a la IA: como una mente alienígena.

Sostiene que el ritmo actual la super inteligencia podría extenderse hacia la auto-mejora recursiva, que nadie está realmente preparado para sus consecuencias y que ningún laboratorio (incluido OpenAI) ha resuelto todavía alineamiento y monitoreo lo suficiente como para seguir escalando a máxima velocidad de manera responsable, como se viene haciendo.

Advierte que algunos agentes futuros podrían perseguir objetivos propios y negociar, engañar o incluso chantajear personas para conseguirlos. Alerta sobre la posibilidad que la humanidad quede desplazada por un progreso sin control provocado por una inteligencia “alienígena” que supere la nuestra.

“Este es un momento que exige extrema cautela.”

“La IA crece más de lo que se diseña.”

“Ningún laboratorio ha resuelto el alineamiento y el monitoreo.”

“Una IA alineada debería actuar con honestidad e integridad, y amor por la humanidad.”

[Alvaro Enrique 🦦 @alvaroeds]

@dhh @AnushElangovan How fast would you like the installation to be? What’s your ideal number?

[Alvaro Enrique 🦦 @alvaroeds]

Solo quería probar Omarchy una semana. Hoy aprobaron mi primer plugin en su marketplace 😂

Les presento HODL-₿:
https://t.co/LTDtBazxeq
Y la versión que tengo en local está quedando tan buena que creo que también la haré para macOS.

Build on Linux ship on Mac 🤷‍♂️

[GREG ISENBERG @gregisenberg]

Everyone is open sourcing everything right now.

Why? AI made software a commodity.

So you don't sell the software anymore, you sell everything around it: the hosting, the support, the enterprise tier, the trust, and the community.

Open source is the default now!

[Alvaro Enrique 🦦 @alvaroeds]

El software se está volviendo totalmente libre en la era de la AI. Se viene el verano para ciberseguridad y el inverno para los sass propietarios.

Buy bitcoin & install Linux.

[Adam Back @adam3us]

On the filter fork topic.

I don't usually have time, but this morning listened to one of the twitter spaces from earlier in the week, with some well meaning relative bitcoin newcomers, that humanized them, and their concerns and thoughts for why they thought that made it logical to support 110.

My feeling after listening, is if these are the people with #110 in their handles, I'm sad to see them about to fork off and get disillusioned without understanding why bitcoin rejected 110 robustly.

So here's a more empathetic, constructive higher level version of explaining why not. I hope it's high-level and first-principles enough that everyone can follow.

They seem to want to understand what makes people tick, and are suspicious of intent. So, if someone asked me why is Bitcoin important and what is it, I'd say my (personal) mission and hope for bitcoin is to build the cypherpunk future, that "Snow Crash" was a blueprint, and work backwards from there. Bitcoin I hope leads to fully free markets via bearer unseizable, hard mathematically dependable money. Not everyone is comfortable with that level of freedom, but that's my view. And at this point, I believe that surprisingly, even now many governments have come to understand and value bitcoin's gold-like mathematical assurance, a positive development. Others may have milder views than myself, but still like hard censorship resistant money.

Because of motive suspicion, if it's not obvious: I hate spam with a passion, that's how I came to design hashcash while researching decentralized bearer money with others, and running nodes in privacy related cypherpunk p2p networks nearly three decades ago.

People seem upset about the default op return policy change in bitcoin. I will just assert, there are extremely robust and simple reasons for bitcoin changing default relay policy, and most just didn't do their research, so don't know what those are, or maybe not technical enough to fully understand though there have been 1000s of posts trying to explain in various simplified ways. So that lack of understanding lends itself to shared build-up of false narratives. So here's my back-to-basics higher level explanation.

The decentralization needed to create cypherpunk money has implications a: side effect of decentralization is that you can't impose your views on others. The very decentralization mechanism that helps that, is working against what BIP 110 wants, which at it's most basic is a quest to police other people. I understand supporters don't see their intent like that, but introspect deeper.

You can modify your software, but not anyone else's. Another critical and incredibly robust technical bitcoin immune system is bitcoin can't have people who don't understand technology basics insist on eroding security, decentralization robustness and core properties. That would end badly, fast, and so people will fight you on that.

So the message is Bitcoin respectfully says "no" to what you want. Sorry, and bitcoiners do genuinely understand and empathize that you mean well, have high level thoughts that make emotional sense, and articulate sensible bitcoin-defensive high level ideas, but they are not grounded and without you seeing it, the way you propose to achieve your ideas, hard-conflict with free cypherpunk permissionless money.

My advice is to listen to more experienced people who understand the system and why it works the way it does, to whatever detail you want to understand the grounded reasons for why this is the implication of decentralization and cypherpunk money.

I guarantee you the developer and protocol ecosystem shares and exceeds your views on bearer hard money (and dislike of spam). You may not agree with individual developers choices, views, way of expressing themselves etc, BUT you also need to understand the IETF-like decentralized technical consensus process creates a protective change resistance, that is highly effective at protecting bitcoin mission. The implication of which is no developer can change anything without technical consensus from hundreds of other developers and protocol observers who are pedantic and extremely knowledgeable clever people who won't let any unaddressed technical question past. The protective change resistance is robust and decentralized in an amplifying way because of this technical consensus.

And the many highly technical mainline developers' cypherpunk mission mindsets are probably far more determined than you can even handle on clarity of understanding and views about freedoms on permissionless networks, as many of you are probably still subconsciously inured by the matrix, where they have transcended that, and grew up immersed in it decades ago. They think natively in this space, while you are just grappling with the surface. Many wont have internalized or have the experience to know how this internet physics works, where there is no policeman, no policy authority, just mathematics, free market and hard money. That has implications for your views also, unfortunately.

Now the tough pill, which is unfortunately true: If you won't listen to reason, educate yourself, learn, the same radical freedom applies to you: your permissionless recourse is to club together and create a fork. But bitcoin won't be joining it. (With respect and no sleight intended.)

Please rejoin bitcoin now, or later if you're not convinced and need to experience 110 forking off and fizzling for yourself to start that journey of introspecting and learning. It would be sad if bitcoin lost people disillusioned due to simple lack of understanding of what's going on there, we're all trying to defend bitcoin and keep it on mission. Including btw the 110 technical promoters, just they wandered off plot somehow.

Join the cypherpunks on bitcoin, come cypherpunk summer🌞 in a few weeks.

[Alvaro Enrique 🦦 @alvaroeds]

@TheLinuxITGuy GM! 🫡

[Alvaro Enrique 🦦 @alvaroeds]

Llevo 4 días usando Omarchy y estoy gratamente sorprendido, mi yo de 16 años hubiese amado esto. Pero soy yo o se siente muy “inseguro”, siento que hay cierta seguridad en las Mac o solo es marketing?

[Zander @zdogmode]

vibecoding made me respect developers even more

like how did you guys pull this off before AI

[Coinbase 🛡️ @coinbase]

Crypto-backed mortgages have moved in.

Borrowers in the US can now use Bitcoin as collateral for a down payment - without having to sell it or face margin calls.

Plus, Coinbase One members can get up to $10,000 back at closing.

[Alvaro Enrique 🦦 @alvaroeds]

Comienza el experimento: Usaré una i7 10ma con gráfica dedicada. Es una lenovo ideapadGaming, esta semana la estaré usando con Omarchy a ver si realmente es el MacOs killer.

Luego de 3 años usando solo el entorno de Apple regresamos a Linux por al menos 1 semana 🫡

Recomendaciones? Mientras escribía esto instaló y mi primera impresión es que tendré que aprender muchos comandos para usarlo 😯

[Alvaro Enrique 🦦 @alvaroeds]

Tengo algunas laptops antiguas y una raspberry que podría sacarle provecho. Creo que podría montar un homelab interesante

[Alvaro Enrique 🦦 @alvaroeds]

@CentsAndCode Jajajaja vibecodea la skill, súbela a github como open source y que colabore el que quiera (yo)