Geometric Bayesian Update — LessWrong