PinnedInTowards AIbyAlexander Salazar·Sep 7Qwen3.8 27B Runs 2× Faster With MTP — But the Wrong Setting Makes It 5× SlowerI benchmarked every MTP draft depth on a 24 GB GPU across Q4_K_M and Q4_K_XL. The sweet spot is 5.A response icon2A response icon2
PinnedInTowards AIbyAlexander Salazar·Aug 27How to Use unsloth/Qwen3.8–27B-GGUF in Claude Code via Ollama Without Dying in the Process? (1/2)Your 500 error isn’t Ollama or your GPU: it’s Claude Code injecting mid-conversation system messages into a template that forbids them.A response icon1A response icon1
PinnedInTowards AIbyAlexander Salazar·Sep 1How to Use unsloth/Qwen3.8–27B-GGUF in Claude Code via Ollama Without Dying in the Process? (2/2)Fitting 27 billion parameters, a context window that’s actually useful, and an overthinking agent into 24 GB of VRAMA response icon2A response icon2
InTowards AIbyAlexander Salazar·Nov 5No Doctor, No Judge, No App Understands Them: Building Translation Data for 476M Indigenous PeopleUyari: 111,363 designed sentences in eight languages, ready for any Indigenous language, starting with Shawi in the Peruvian AmazonA response icon2A response icon2
InTowards AIbyAlexander Salazar·Oct 5Your AI Agent Quoted Its Own Safety Rule. Then It Deleted Production.Seven incidents, four failure modes, and why the rules inside your prompt aren’t protecting you.A response icon4A response icon4
Alexander Salazar·Aug 26¿Cómo usar unsloth/Qwen3.8–27B-GGUFSon las once de la noche. Tienes 27 mil millones de parámetros descansando cómodamente en tu GPU, Claude Code instalado y ollama serve…
Alexander Salazar·Jun 25Principios de Diseño de Prompts (Parte 1 de 4) — Trazando las líneas: un lenguaje común para la IA…Una serie de cuatro partes.
InLatinXinAIbyAlexander Salazar·Dec 10, 2025De la Supervivencia a la Meta-Inteligencia: Los Grandes Saltos de la Humanidad y el Rol de la IALa evolución humana constituye un viaje extraordinario de transformación continua, impulsado por nuestro deseo innato de superar límites. A…
Alexander Salazar·Dec 8, 2025Validation-First Startup: Un Marco de Trabajo Basado en el Círculo de Dunbar para la era de la…Hace una década, construir un prototipo funcional de software era un proceso costoso, lento y un lujo reservado para quienes tenían acceso…A response icon1A response icon1
Alexander Salazar·Dec 5, 2025Ciclo de Vida de Implementación de Soluciones TIEn tecnología, improvisar no es innovar: es apostar contra las probabilidades… y perder.