Recent discussions among researchers highlight the limitations of Large Language Models (LLMs) in reasoning and planning. Yann LeCun stated that models like OpenAI's o1 primarily perform intelligent retrieval rather than demonstrating true reasoning capabilities. A new paper titled 'CPL: Critical Planning Step Learning Boosts LLM Generalization in Reasoning Tasks' aims to enhance LLMs' generalization in reasoning tasks by teaching them to recognize dead-ends in their thinking processes. However, experts like Gary Marcus emphasize that LLMs lack formal reasoning abilities, which presents a significant challenge. Other researchers have noted that while LLMs can mimic reasoning through pattern matching, they do not possess a genuine understanding of logical relationships. The debate continues on how to improve LLMs' reasoning skills, with some suggesting that tree search methods could yield better results than traditional reasoning approaches. As the field progresses, the effectiveness of LLMs in solving complex problems remains a focal point of research.