Imagine a world where designing new drugs is as easy as writing a description. Large language models (LLMs), the powerhouses behind AI chatbots, are showing potential in revolutionizing drug discovery. However, crafting molecules with specific properties, like those needed for effective medicines, has remained a formidable challenge. New research tackles this hurdle with a framework called PEIT (Property Enhanced Instruction Tuning) that empowers LLMs to not only understand molecular structures and textual descriptions but also, crucially, link them to desired biochemical properties. This two-step process involves first pre-training a model (PEIT-GEN) to align these different data types, and then fine-tuning a powerful open-source LLM with this newly acquired knowledge. The result? A PEIT-LLM that can generate molecules tailored to specific requirements. Tests show PEIT-LLM excels at tasks like molecule captioning (describing a molecule's structure in words) and, most excitingly, multi-constraint molecule generation where it designs molecules that satisfy a list of desired properties. This marks a significant step towards streamlining the complex drug development process, holding promise for creating more effective and targeted therapies faster. While challenges remain, such as incorporating more diverse data types like molecular images and automating aspects of the process, this research opens exciting new possibilities for AI-driven drug discovery, potentially transforming the pharmaceutical landscape.