The SIA population update can be surprisingly small — LessWrong