In contrast, a study by researchers from New Zealand and the US examined ChatGPT-4o and a Custom GPT on first-year civil engineering statics problems. Unlike their performance on theoretical physics challenges, the AI models struggled with nuanced engineering questions—such as distinguishing tension from compression and resolving inclined forces—and exhibited limited proficiency in interpreting diagrams. These shortcomings highlight that real-world engineering relies on iterative, collaborative workflows—peer review, simulations, and safety factors—rather than isolated, purely theoretical problem-solving. https://arxiv.org/abs/2502.00562
PS - A more impactful application of ChatGPT in civil engineering lies in assisting with compliance with the strict codes and standards that govern design projects, ensuring they meet required levels of safety, reliability, and efficiency. Checking that every design follows all the detailed regulations can be really complex and time-consuming—it’s usually something expert engineers do manually. A recent paper introduces an open-source framework that uses large language models (LLMs) to provide engineers with accurate, code-related answers from natural language queries, including references to relevant code sections. The initial implementation, based on the National Building Code of Canada, shows promise demonstrating the potential of this approach to streamline design verification tasks. https://ascelibrary.org/doi/10.1061/JCCEE5.CPENG-6037#con1
