Imagine an AI doctor fluent in Portuguese, ready to answer medical questions. That's the goal of a recent study exploring how Large Language Models (LLMs) can be adapted for healthcare in Portuguese-speaking communities. Researchers experimented with fine-tuning a model called ChatBode-7B using translated medical datasets. They wanted to create a virtual medical assistant, a chatbot specializing in providing medical information in Portuguese. One challenge was the lack of a dedicated Brazilian Portuguese medical conversation dataset. Existing options were either outdated or lacked professional medical verification. As a workaround, the team translated English datasets using GPT-3.5, aiming for accuracy in medical terminology. They tested different fine-tuning approaches using datasets like HealthCareMagic-100k-en and MedQuAD. Interestingly, the InternLM2 model, which had prior training on medical data, performed the best in terms of accuracy, completeness, and safety. However, the DrBode models, derived from ChatBode, experienced what's called 'catastrophic forgetting'—losing some previously learned medical knowledge during the new training process. Despite this, DrBode models excelled in grammaticality and coherence of their responses. Evaluating the models presented another hurdle: low agreement among medical professionals who rated the responses. This highlights the need for better evaluation methods in this specialized field. A key concern raised was the models' tendency to recommend medications even when not appropriate, like suggesting ibuprofen for dengue fever symptoms – a potentially dangerous recommendation in Brazil where dengue is common. This underscores the importance of training models on data reflecting regional health contexts. This study represents a significant step toward creating AI-powered medical assistants for Portuguese speakers. Future research will focus on developing native language datasets, improving fine-tuning to prevent knowledge loss, and creating more robust evaluation methods. This work has the potential to improve access to reliable medical information and enhance healthcare outcomes for Portuguese-speaking populations worldwide.