From Finite Factors to Bayes Nets — LessWrong