Cost-effectiveness of professional field-building programs for AI safety research — LessWrong