Maxent and Abstractions: Current Best Arguments — LessWrong