Miguel Garcia

In pursuit of a maximally predictive world model. Current rabbit holes include geometric deep learning, statecraft, and ethnography.

May, 2025

  • Currently building narindex. An attempt to gain proficiency in the more practical aspects of deep learning. In a sense, reconciling too much paper reading with something more visceral. For some time, I have reflected on how GDL as a framework could be used for program synthesis tasks—more on this later. Yet nar_index, I believe, is the first step towards that.

    • Also, I have found MLX incredibly delightful. Never thought a Mac Mini could be capable of so much. Started working on developing custom Metal kernels for common GNN operations.

  • Inspired by Fukuyama's approach in "The Origins of Political Order", I have developed an interest in studying state capacity from an ethnographic approach. That is to say, an attempt to derive explanations for societal success and/or failure in terms of the interplay between institutions and innate—so to speak—human cultural tendencies. Still thinking about this, but perhaps will release a proto-essay soon.

Selected Writings