Sunday Reads #6

• I love Maro and I love Harry Styles.
• For the last couple of days I’ve been binge-listening to Fashion Sucks. While I’ve never been particularly interested in fashion (though I wish I were), I’m fascinated by understanding how industries work from the inside and cracking their cultural codes. The hosts, who are fashion industry insiders, bring on Spanish stylists, designers, and other professionals to discuss their work and the nuances of their roles. What I particularly appreciate is how they’re addressing an issue they often highlight: the lack of recognition in Spain for those who achieve success internationally. Don’t miss their interview with Alejandro Palomo and Harry Styles performing As It Was in that iconic Palomo Spain look.
• I have just finished Nocturno de Chile. I started it because I had never read Bolaño and a 130-page book seemed like a small investment. The book felt longer than expected, I guess because it’s written as a single continuous paragraph (stream of consciousness-like).
• I am midway through Ezra Klein’s Abundance. This book addresses what I think is the most important cultural problem of our time: it’s not enough to redistribute, we need to build more and better. I hope this gets translated soon.
• Mistral at Infra @ Scale.tl;dw: production is hard. The real moat isn’t model size but operational resilience. You need solid monitoring and automated tests because even a minor GPU driver change can tank latency. Inference infrastructure (rate-limiters, autoscaling, circuit breakers) takes more engineering time than training (this resonates a lot). Customer diversity (banks wanting air-gapped clusters, media firms chasing latest checkpoints, multilingual clients needing tokenizer tweaks) means your deployment stack needs to be modular and configurable. The Stanford class more generally is quite good, they had Guillermo Rauch and Ben Mann before.
• Javi sent me this yesterday. What struck me most about the article was the real portrayal of the learning process. True learning is inherently painful (yes, painful) - it stretches your mind in uncomfortable ways. When mentoring people, there’s an instinct to spare others the struggle by sharing everything you’ve learned (at least I have it), but that would rob them of developing the most crucial skill: the ability to learn and adapt independently. You might need to do something like this.
• Last but not least, I’ve decided to read up on some historical AI papers. The first one is A Logical Calculus of the Ideas Immanent in Nervous Activity. This is just meant to be some of my notes that help me understand what I am reading, just like anything on this blog: I write primarily to understand things myself. You can follow along with me here.