I come from NeuroAI [1, 2] and moved downstream to efficient foundation model design, training and inference.
I open-sourced the Maths, CS & AI Compendium, which I gradually wrote over the past 7 years. Then I co-founded Cactus Compute, a Y Combinator company where some of my personal works include Hadamard MLP, Intelligence Ladders, Simple Attention Networks and Cactus Needle, plus most of Cactus Kernels. I also manage the research, engineering, direction and team behind Cactus Whistle, Cactus Quants and Cactus Hybrid.
On paper I'm small by AI researcher standards, with a master's and no badges that impress. However, impact is like a shadow on the wall, and sometimes a "small" man casts a large shadow (Alec Radford, Georgi Gerganov, etc.). I wanted a way to find those people among Cactus applicants and "de-slop" the pile, without leaning on the same badges that would have filtered me out. So I wrote R² Index to rank people by what they've actually produced, from measurable impact.
But something surprising happened. It rediscovered AI lore on its own. The godfathers of AI rank top three of all time, and the top universities and VCs come out where you'd expect. That made me trust it enough to share. For the record, I don't even make the top 20k of these ~800k researchers. My co-founder Roman Shemet chose the name!