## Aligning Language Models with Chinese Medical Ethics A new study introduces MedES, a system for aligning large language models (LLMs) with the nuances of Chinese medical ethics. The system is based on a benchmark created from 260 authoritative Chinese sources in the medical, ethical, and legal fields. ## MedES: A Dynamic Benchmark MedES is designed to reflect the challenges of clinical decision-making. To facilitate model alignment, a framework has been introduced that leverages a dedicated automated evaluator, trained on expert-labeled data and with over 97% accuracy. ## An Innovative Approach This evaluator generates targeted prompts and provides structured ethical feedback. Using this pipeline, a 7 billion parameter LLM was improved through supervised fine-tuning and domain-specific preference optimization. ## Experimental Results Experimental results, conducted entirely within the context of Chinese medical ethics, demonstrate that the aligned model significantly outperforms larger baseline models in core ethical tasks, with observed improvements in both quality and composite evaluation metrics. This work offers a practical and adaptable framework for aligning LLMs with medical ethics in the Chinese healthcare context and suggests that similar alignment pipelines can be instantiated in other legal and cultural environments through modular replacement of the underlying normative corpus.