Imagine a world where scientists can design molecules with specific properties on demand, revolutionizing medicine, material science, and chemistry. This vision gets closer to reality with a novel approach by using large language models (LLMs) for molecular generation. Traditionally, molecular generation has relied on rule-based systems or complex graph-based models, but these often struggle to produce diverse and valid chemical structures. Large language models like those powering ChatGPT excel at understanding text, but molecules are typically represented as complex graphs, not words. This mismatch poses a significant challenge for applying LLMs to molecular design. Researchers have developed a clever solution: transforming molecular graphs into tree-structured text formats like JSON and XML. These tree-like formats preserve essential information about atoms and bonds and are something LLMs are already trained to interpret. This innovation, called G2T-LLM, uses a "graph-to-tree" encoding, converting intricate molecular graphs into hierarchical text that LLMs can easily process. Once the molecules are in this text format, the LLM is fine-tuned with a "molecular completion" task. It's given a partial molecule and has to predict the rest, learning chemical rules and constraints in the process. The results are impressive. G2T-LLM achieves state-of-the-art performance in generating valid, diverse, and novel molecules. While not quite ready to replace expert chemists, it empowers researchers to design molecules tailored to specific properties. This breakthrough has exciting implications for drug discovery, creating new therapies, and developing novel materials. The ability to generate made-to-order molecules with specific characteristics accelerates research across various scientific domains.