John D. Cook • 3/4/2026

From logistic regression to AI

The article contrasts classical statistical models like logistic regression with modern large-scale neural networks and LLMs. It discusses parameter efficiency, data requirements (like the 10 events per parameter rule), and how over-parameterization behaves differently in classical vs. neural network contexts. The author uses examples from clinical trials and patent applications to illustrate the shift from modest, carefully-parameterized models to the 'black art' of developing billion-parameter models that achieve amazing results.

0 comments

#Logistic Regression #Neural Networks #Bayesian Statistics