Large Language Models (LLMs) are revolutionizing fields from writing to coding, but can they truly grasp the complexities of chemistry? While general-purpose LLMs like GPT-4 display impressive abilities, they often stumble when faced with chemical concepts, molecular structures, and the nuances of chemical reactions. This isn't surprising, given that they're primarily trained on web text, which contains limited chemical information. A recent research survey explores how to transform these generalist LLMs into specialized chemical experts. Researchers are tackling three key challenges: imbuing LLMs with sufficient chemical domain knowledge, enabling them to perceive and process multimodal chemical data (like 2D graphs, 3D structures, and spectra), and equipping them to utilize existing chemistry tools and databases. One approach is continued pre-training on massive chemical datasets, including scientific papers, textbooks, and molecular databases. Supervised fine-tuning (SFT) then tailors the LLM to specific chemical tasks like predicting reaction outcomes or designing molecules with desired properties. Reinforcement learning from human feedback (RLHF), a technique that uses human preferences to guide model behavior, helps to refine the model's responses and minimize errors or "hallucinations." However, chemistry extends beyond text. Researchers are developing innovative ways to integrate multimodal information. Specialized encoders transform molecular structures and spectral data into formats that LLMs can understand, and novel alignment techniques help bridge the gap between these different modalities. Finally, LLMs are being linked to external chemistry tools, databases, and even robotic systems. This allows them to access up-to-date knowledge, perform complex computations, and even autonomously conduct experiments. The development of robust benchmarks is also crucial for assessing LLM performance on various chemical tasks. Benchmarks like ChemLLMBench and SciKnowEval evaluate LLMs' knowledge coverage, reasoning abilities, and practical applications. While significant progress has been made, the journey toward creating a truly intelligent chemical LLM is ongoing. Future research will focus on creating more diverse training datasets, improving multi-modal alignment, developing more effective RLHF strategies, and exploring innovative applications like automated experimentation and personalized chemical research assistants. The potential for AI to transform chemical research is immense, offering a future where LLMs accelerate discovery, design new materials, and ultimately deepen our understanding of the molecular world.