“Current Large Language Models cannot perform System 2 thinking, which involves deliberate planning and devoting more computational resources to more complex problems.”