Imagine an AI that can tap into vast knowledge bases to answer questions accurately without extensive training. That's the exciting premise of the latest research in "zero-shot reasoning" using knowledge graphs. Large Language Models (LLMs) are amazing at generating human-like text, but they often struggle with factual questions or tasks requiring specific knowledge. Knowledge graphs, structured databases of information, offer a solution. The challenge is efficiently combining these two powerful tools. The "Tree-of-Traversals" algorithm offers an innovative method to bridge this gap. It enables an LLM to navigate through a knowledge graph like an expert detective, gathering relevant information step-by-step. This navigation is guided by a "tree search" that explores different reasoning paths, constantly evaluating and refining its search. The algorithm uses clever prompts to guide the LLM’s actions, avoiding the need for explicit training data. This allows the LLM to ask the knowledge graph questions and gather facts dynamically. This process repeats until the LLM reaches a conclusion with a high confidence score. Researchers tested this approach with different LLMs and datasets, including a new one combining general knowledge with domain-specific music information. Results showed Tree-of-Traversals significantly outperforms existing methods, particularly in complex multi-hop reasoning. In one compelling example, the algorithm correctly determines which musical venue was opened more recently by navigating both Wikidata and MusicBrainz. The zero-shot nature of Tree-of-Traversals also allows companies to connect their LLMs to proprietary knowledge bases without model retraining. While this technology has the potential to unlock new capabilities, the authors highlight important ethical considerations such as the effect on AI safety and the need for cross-language evaluation. The combination of LLMs and knowledge graphs offers a powerful approach to knowledge-intensive reasoning tasks, and Tree-of-Traversals provides a promising path towards more accurate and reliable AI systems.