Abstract: Large Language Models (LLMs) typically reason via Chain-of-Thought (CoT) prompting or explicit training. Though many LLMs achieve similar accuracy on challenging tasks, such as math problem ...