“Joon Sung Park believes the ability of current large language models to simulate human behavior has plateaued because they are optimized for rationality, whereas humans are often irrational.”