It is impossible to disentangle technical innovation from technical debt. Machine learning feels theoretical, but in practice it runs on benchmarks, leaderboards, and social norms. In this post, Ben Recht reflects on why ML progress is driven more by engineering culture and empirical testing than clean theory. And why that tension is impossible to escape.

What I learned when I stopped trying to make life exciting

A technique for deterministic labeling from stochastic models, with benchmarked Golang implementation.