The field of medical AI is rapidly advancing, with a focus on improving clinical decision support and error correction. Recent research has highlighted the importance of developing large language models (LLMs) that can accurately capture domain-specific knowledge and notation, particularly in high-stake applications such as medical diagnosis and treatment. Notably, the development of benchmarks and evaluation frameworks has enabled the assessment of LLMs' performance in various medical tasks, including medical order extraction, error correction, and medication safety. These benchmarks have revealed areas where LLMs struggle, such as in handling contraindication and interaction knowledge, and have provided insights into improving reliability through better prompting and task-specific tuning. Furthermore, the introduction of novel frameworks and architectures, such as multi-agent systems and reinforcement learning environments, has shown promise in enhancing pre-consultation efficiency and quality in clinical settings. Overall, the field is moving towards developing more accurate, reliable, and transparent medical AI systems that can support clinicians in providing high-quality patient care. Noteworthy papers include MedCalc-Eval and MedCalc-Env, which introduce a comprehensive benchmark and environment for evaluating and improving LLMs' medical calculation abilities, and RxSafeBench, which provides a comprehensive benchmark for evaluating medication safety in LLMs.