New paper is out! Poor Alignment and Steerability of Large Language Models: Evidence from College Admission Essays