Researchers have developed a new method called FlightLLM to analyze flight safety events. The approach uses large language models to identify the underlying causes of risk events, such as hard landings, and provides clear explanations at the level of pilot control behavior. To address challenges like modal inconsistency and limited data, the team combined statistical descriptors with physically meaningful flight indicators and incorporated a statistical expert called CatBoost to improve classification performance. The method was evaluated on 704 real-world A320 flight samples and achieved competitive results while generating direct and reasonable explanations for event causes. --- Why it matters: This work matters because it provides a promising solution to the challenge of interpreting flight safety events, which is crucial for improving aviation safety. By using large language models to generate clear explanations at the level of pilot control behavior, researchers can better understand the underlying causes of risk events and develop targeted interventions to prevent them. Source: https://arxiv.org/abs/2608.18017