Machine-Readable Prevalence Estimates — LessWrong