In my last post, I referenced Dave Friedman’s analysis of the Aschenbrenner manifesto highlighting research challenges in data efficiency improvements — how to use less data for equivalent training.