A recent study published on arXiv investigates the reasoning biases of next-token predictors in large language models, focusing on their tendency toward systematically inefficient problem-solving strategies. The research highlights the importance of systematic and incremental reasoning, noting that while increasing computational resources can help address complex tasks, it may also introduce redundancy in the reasoning process. The study employs a shortest-path case study to analyze how structured chains of thought can improve problem-solving efficiency. This approach draws an analogy to human cognitive processes, suggesting that organized, step-by-step reasoning enhances performance. The findings contribute to ongoing discussions in natural language processing about optimizing reasoning capabilities in large language models. By emphasizing structured reasoning, the research points toward potential improvements in how these models tackle complex problems. This work aligns with broader efforts in the field to refine model reasoning and reduce inefficiencies.