Hi, I’m Ed. By day I’m a Senior Scientist at Zscaler, working at the intersection of cybersecurity and AI. By night (and sometimes by day, if I can get away with it) I think about a single question:
What does it take to build intelligent systems out of smaller, reusable parts, and why does the field keep tripping over itself when it tries?
The words I keep coming back to are modularity and compositionality. In plain terms: can we build AI the way we build pretty much everything else that works, which is by snapping together well-understood pieces? In less plain terms: when do modules emerge inside trained networks, when do learned primitives compose into genuinely new behavior, and what are the right interfaces between parts at each level of the stack?
This site is my working notebook. The Research page lays out the program in more detail and collects longer posts. Notes is where shorter, rougher thoughts live. Quotes is a personal page of lines other people said better than I ever will.