Logarithms

Research

We study how specialist models are trained — rich human-generated data, reinforcement learning against one domain, continuous evaluation — and how they compose into swarms that improve themselves.