Edu Mínguez

🏠 Home: eduardominguez.es ✍️ Blog: underkube.com 💬 Mastodon: @minWi" target="_blank" rel="nofollow noopener">tty0.social/@minWi 🟦 Bluesky: bsky.app/profile/minwi.…

Huh, so that's why text is called a string

For my friends who are still using UV and might be a little weary about recent compromises to PyPi packages, stick this in your pyproject.toml. You can let all of those pip users find and report the compromises...

If you use GitHub (especially if you pay for it!!) consider doing this *immediately* Settings -> Privacy -> Disallow GitHub to train their models on your code. GitHub opted *everyone* into training. No matter if you pay for the service (like I do). WTH github.com/settings/copil…

when you are vibe coding and you finally try the app after an hour

@thdxr Just add 'Seriously.' at the end of the sentence (the dot is important). It works all the time. Seriously.

any tips on getting your team to take your suggestions more seriously?

If @Metallica did the weather 🤘 🎸

TurboQuant might be one of the most important efficiency breakthroughs in LLM inference lately. Here’s a quick breakdown 🧵

The older I get, the more I understand this guy.

Iniciando sesión en cualquier web en 2026.

underkube.com/posts/2026-03-… My weekly reading list is live. This week features deep dives into Btrfs enterprise scaling and cost reduction, a flurry of new open source AI agent control planes, and critical security updates for CI/CD pipelines.

@456c6f727269 Check gitlab.com/IsolatedOctopi…

A developer managed to run a massive Qwen 3.5 397B locally on a MacBook by streaming model weights from the SSD instead of loading them into RAM, based on LLM in a Flash. Thanks to Apple’s fast storage and unified memory, plus MoE efficiency (only a small fraction of the model is used per token), it achieves usable speeds (~5–7 tok/s) with minimal RAM. Even more striking: the entire system (inference engine, GPU shaders, optimizations) was written by Claude Opus 4.6 via an agentic workflow.

An iconic image of the Space Shuttle Endeavour being transported by a Boeing 747 as it flies over the Mojave Desert during its transfer from California to Florida on December 10, 2008. 📸: Carla Thomas

@minWi Mi homelab lo componen 1 minipc de segunda mano de 100€, 1 minipc chino de 140€ y una torre que compré en 2011... 😅

👀

@456c6f727269 Yo es que soy del puño _cerrao_ para homelab 😁 Tiene pintaza el GMKtec también!

@minWi Espero que este tipo de equipos se vayan generalizando y bajen de precio. De momento, le tengo el ojo echado al GMKtec EVO-X2.

Cuando vuelve a casa, sus amigos peces nunca le van a creer🐟💪

If you love self hosting, local AI, and technical deep dives, this week's reading list is for you. Turning an old Galaxy S10 into a server, an open OS for autonomous agents, and a Rust based CLI proxy to reduce LLM token consumption. underkube.com/posts/2026-03-…

The "Severance" workflow

@zetalyrae - flexibility - openness - choice - standardization - performance

Areas where Linux is better than macOS: - systemd much better than launchd - sandboxing (not deprecated unlike sandbox-exec) - package management - declarative config management - obviously customization