To expose the limitations in the reasoning of LLMs