ABOUT
From Richard Sutton's The Bitter Lesson: across AI history, general methods that can absorb more computation have ultimately pulled far ahead of hand-designed systems. Search and learning turn additional compute into capability; human knowledge may help in the near term, but does not offer the same long-run curve.
Foundation models still contain many hand-designed choices, from the architecture and training harness to activation functions. Searching broadly across those choices is prohibitively expensive, and there is no broadly accepted smaller unit to search and then scale up. That is why Pyco was founded.
Pyco was started by Seth Rothbaum, an ML researcher at the University of Pennsylvania.
Reach out at seth@pyco.ai.