Researchers have evaluated the ability of large language models to automatically repair errors in planning domain definition languages (PDDL) using a new approach. They compared the performance of symbolic baseline methods with open-weight large language models and found that the latter achieved better results but still struggled to guarantee reliable automated model repair. The best-performing LLM reached an F1 score of 0.87, but its mean test pass rate was only 0.82, indicating that current models are not yet reliable for this task. --- Why it matters: This matters because it affects the reliability and trustworthiness of AI planning systems, which rely on accurate models to make decisions. Improving the ability to automatically repair errors in these models is crucial for developing more robust and efficient AI systems. Source: https://arxiv.org/abs/2608.17341