Imagine an AI assistant that could guide scientists through the intricate world of materials synthesis. Researchers put this idea to the test by evaluating a large language model (LLM) on its knowledge of atomic layer deposition (ALD), a crucial technique for creating thin films in microelectronics and energy applications. ALD involves depositing materials layer by atomic layer, offering precise control over film properties. Think of it like building with atomic LEGOs. This new benchmark, called ALDbench, challenged the LLM with 70 open-ended questions ranging from basic graduate-level inquiries to complex queries demanding expert-level understanding. Human experts graded both the questions and the LLM’s answers on criteria like difficulty, specificity, relevance, and accuracy. The LLM achieved a respectable average score, similar to a passing grade. However, it stumbled on nearly 40% of the questions, revealing a tendency to 'hallucinate' or fabricate answers, particularly when dealing with specific chemical precursors. Interestingly, the LLM performed best on questions requiring less specific answers, suggesting a current limitation in its capacity to provide precise, detailed information. While this research highlights the ongoing challenges in applying AI to scientific domains, it also provides valuable insights into how to further refine LLMs for complex technical tasks. Future research will explore techniques like prompt engineering and hyperparameter tuning to enhance LLM performance in materials science, paving the way for AI-powered tools that could accelerate discovery and innovation in this critical field.