January 2026

Ten years ago this week, I wrote a provocative and bold post that blew up, made it to top spot on HackerNews. I had just joined Magic Pony, a nascent startup, and I remember the founders Rob and Zehan scolding me for offending the very community we were part of, and – of course – deep learning developers we wanted to recruit. Deep Learning is Easy – Learn Something Harder Caveat: This post is meant address people who […]

Ver mais

Over winter break I built a prototype which is effectively a device (currently Raspberry Pi) which listens and detects “meaningful moments” for a given household or family. I have two young kids so it’s somewhat tailored for that environment. What I have so far works, and catches 80% of the 1k “moments” I manually labeled and deemed as worth preserving. And I’m confident I could make it better, however there is a wall of optimization problems ahead of […]

Ver mais

Understanding VL-JEPA and its approach to embedding-based vision–language modeling Modern vision language models can describe images, answer questions, and interpret videos with impressive fluency. Yet they all share an unusual habit: they talk constantly. Even when nothing meaningful has changed. Even when the model already understands what is happening. This is not a stylistic flaw it is a structural one. Most vision–language models (VLMs) are built on the same foundation as large language models: token-by-token generation. They must […]

Ver mais

About a decade ago, many media outlets—including WIRED—zeroed in on a weird trend at the intersection of mental health, drug science, and Silicon Valley biohacking: microdosing, or the practice of taking a small amount of a psychedelic drug seeking not full-blown hallucinatory revels but gentler, more stable effects. Typically using psilocybin mushrooms or LSD, the archetypal microdoser sought less melting walls and open-eye kaleidoscopic visuals than boosts in mood and energy, like a gentle spring breeze blowing through […]

Ver mais

Plantear si existe una relación entre determinadas trayectorias biográficas y los problemas estructurales de la tecnología contemporánea no es un ejercicio de culpabilización retrospectiva, sino un intento de comprensión. La tecnología no surge en el vacío: la diseñan personas concretas, con nombres, apellidos e historias concretas, socializadas en contextos políticos, culturales y morales que dejan huella. Ignorar ese hecho es una forma cómoda y profundamente interesada de despolitizar sistemas que hoy ejercen un poder extraordinario sobre nuestras vidas. […]

Ver mais