A study led by scholars at Stanford University and Northeastern University has shed light on why large language models often produce narrow, repetitive outputs and offers a promising technique to ...