Understanding driving risks for older drivers with large language models

Understanding driving risks for older drivers is a critical challenge for traffic safety. Recent advances in multimodal large language models (LLMs) raise the possibility that such models may support scene-level interpretation beyond conventional object detection. However, little is known what extent current LLMs can emulate the integrated, human-like judgments required for older driver diagnostics. Here we show that multi-shot prompting enables a LLM to label static dashcam images with accuracy