About

I'm Marlon Ávalos Elizondo. Two things interest me in equal measure, and they usually travel apart: machines that decide, and the texts that try to set limits on them. I program for pleasure and read regulatory frameworks out of curiosity.

Caja de Arena comes from there.

It's a personal laboratory of technology and artificial intelligence. It doesn't work exactly like a blog, because its basic unit isn't the article, but the experiment: a question, a hypothesis, a method, a test, a cost, and often, something that went wrong.

There's a curious divide in the AI world: many of the people trying to regulate it have never built with it, and many of those building it rarely stop to read the frameworks meant to govern it. That gap strikes me as the most interesting place to look from, and it runs through almost everything I publish here.

I also document the failures. An experiment that doesn't work can be more useful than a flawless demonstration if it makes clear what was attempted, why it failed, and how much it cost to discover. If that log saves someone two weeks of repeating the same errors, it's done its job.

That's why experiments don't get "corrected" either. They get repeated.

Each one keeps a permanent identifier. If months later I return to the same problem, I don't rewrite the previous result: I publish a new measurement. This way, Caja de Arena doesn't aim to build a collection of definitive versions, but a longitudinal record of how technologies, tools, and my own hypotheses about them change.

Everything published here is personal. It does not represent the position of any organization I work with or collaborate with.