Technical AI Safety Roles at Open Philanthropy — LessWrong