Enhancing Cognitive Functions in Large Language Models Towards AGI: A State-of-the-Art Survey and Exploratory Review

This paper presents a comprehensive state-of-the-art survey and exploratory review of current methodologies aimed at enhancing the reasoning capabilities of Large Language Models (LLMs) on the path toward Artificial General Intelligence (AGI). The paper presents a novel synthesis of enhancement strategies across four complementary cognitive dimensions: planning, memory, world modeling, and neuro-symbolic integration. Although LLMs have shown remarkable performance in language-based tasks, their reasoning abilities remain limited, particularly in areas such as causal inference, long-term planning, and common sense understanding, capabilities vital for domains like satellite mission engineering design and spacecraft operations. This paper systematically examine these cognitive bottlenecks and the challenges posed by hallucinations, which undermine trust and reliability. The review then explores promising strategies for augmenting LLM capabilities, including integration with knowledge graphs and symbolic reasoning systems, the development of internal "world models," memory and attention-based architectural innovations, and advanced training techniques such as instruction tuning, reinforcement learning from human feedback, and chain-of-thought prompting. Furthermore, we analyze hybrid neuro-symbolic approaches and multi-agent architectures as emerging paradigms that blend deep learning with structured reasoning. Overall, this work aims to provide a foundation for advancing LLMs toward more robust, generalizable, and cognitively grounded artificial intelligence systems by mapping the landscape of current research and identifying the most promising directions for future development.