国内首个“主任级AI医生”诞生,夸克健康大模型通过12门主任医师考试
Guan Cha Zhe Wang·2025-07-23 06:32

Core Insights - Quark Health's large model has successfully passed the written assessment for chief physician in 12 core medical disciplines in China, becoming the first large model to achieve this milestone [1] - This achievement follows the model's earlier success in passing the associate chief physician examination in May, indicating significant advancements in complex medical reasoning tasks [1] - The model's capabilities are now fully integrated into Quark's AI search, allowing users to access "chief-level AI doctor" capabilities for health inquiries [1] Model Development - Quark Health's algorithm head, Xu Jian, stated that the model is based on Tongyi Qianwen and follows a deep engineering route aimed at vertical scenarios, focusing on training the AI to understand medical thinking rather than just answering medical questions [1][2] - A key breakthrough of the model is the development of "slow thinking ability," which combines chain reasoning and multi-stage clinical deduction modeling, enabling the model to derive answers through phased and in-depth reasoning for complex medical issues [1][2] Engineering System - To build the slow thinking capability, Quark employed a "dual data production line + dual reward mechanism" engineering system, categorizing medical data into "verifiable" and "non-verifiable" types for diagnostic and health advice tasks respectively [2] - The training methodology includes a "process reward model" and a "result reward model" to evaluate the reasoning chain's validity and the accuracy of final conclusions, enhancing clinical interpretability and reasoning consistency [2] User Engagement - Quark Health's large model is supported by a professional annotation team of over 1,000 physicians, with more than 400 being associate chief physicians or higher, ensuring high-quality data input [2] - The Quark AI search platform has attracted a significant user base among medical students and doctors, with monthly active users exceeding 2 million among medical students, representing over half of the demographic [2]