Imagine navigating a vast labyrinth of information, seeking answers hidden deep within its interconnected pathways. That's the challenge of multi-hop question answering over knowledge graphs, where unearthing the truth requires connecting multiple pieces of information scattered across a complex network. Traditional methods, like semantic parsing and information retrieval, have grappled with this complexity, but now, large language models (LLMs) are stepping onto the scene, offering a fresh perspective. This research delves into how LLMs are revolutionizing multi-hop question answering. Instead of relying solely on pre-programmed rules, LLMs leverage their vast knowledge and language understanding to reason through the connections within a knowledge graph. The researchers explore two distinct approaches: one that feeds the LLM relevant snippets of the graph (IR-LLM) and another that provides only the graph's schema (SP-LLM). Interestingly, the best approach depends on the specific knowledge graph. For graphs with readily available sub-graphs, like WebQSP and MetaQA, the IR-LLM method shines, outperforming traditional techniques by a significant margin. In these cases, providing the LLM with contextual information boosts its ability to connect the dots. However, when dealing with massive knowledge graphs like DBpedia or Wikidata, where providing the entire context becomes impractical, the SP-LLM approach takes the lead. By giving the LLM a blueprint of the graph's structure, it can intelligently generate queries to pinpoint the answers. This research highlights the adaptability of LLMs, showcasing their ability to tailor their approach based on the unique challenges of each knowledge graph. While LLMs demonstrate remarkable progress, challenges remain. Ensuring the accuracy and reliability of LLM-generated answers is crucial, especially in critical applications. Future research will likely focus on refining these techniques, improving the efficiency of knowledge retrieval, and developing methods to verify the validity of LLM-generated responses. The potential of LLMs to unlock the wealth of information hidden within knowledge graphs is immense, paving the way for more intuitive and powerful knowledge discovery systems.