Text this: Large Language Models Provide Accurate but Potentially Unsafe Answers to Multimodal Critical Care Medicine Board Review Questions.