It is our vision to develop and evaluate multimodal LLMs that can flexibly ingest medical data modalities and that can be instructed to follow a variety of tasks.