Conditioning on Observers — LessWrong