Quark Health’s Large Language Model Passes 12 Physician-in-Chief Exams – A First in China

CNBC AI News, July 23rd – In a groundbreaking achievement, the Quark Health Large Model (LLM) has successfully passed the written evaluations for 12 core medical disciplines at the Chief Physician level in China. This marks the Quark LLM as the first domestic AI model to conquer this rigorous challenge.

This “Chief Physician-level AI doctor” capability is now fully integrated into Quark’s AI search functionality. Users can access this advanced diagnostic assistance by opting for the “deep search” feature when querying health-related issues.

This milestone comes less than two months after the Quark Health LLM surpassed the passing grade in 12 National Associate Chief Physician qualification exams on May 27th, demonstrating a rapid and significant advancement in its capabilities. The speed of this progress underscores the potential of AI in revolutionizing healthcare.

Performance analysis indicates that the Quark Health LLM exhibits a performance curve where its advantage over general-purpose language models becomes increasingly pronounced as the complexity of the medical challenge escalates.

Built upon the foundation of Tongyi Qianwen, the Quark Health LLM has embarked on a path of deep engineering tailored for vertical applications in healthcare. This strategic choice reflects a focus on specialization and domain expertise, critical for effective AI integration into medical practice.

According to Xu Jian, head of Quark Health’s algorithm development, “We are not merely training AI to answer medical questions, but to cultivate its medical reasoning capabilities.” This approach prioritizes mimicking the cognitive processes of experienced physicians, enhancing diagnostic accuracy and treatment planning.

A core breakthrough in this iteration of the Quark Health LLM is the development of “slow-thinking” abilities. This function refines the model’s ability to handle critical real time decisions.

This capability integrates chain-of-reasoning with multi-stage clinical reasoning pathway modeling, empowering the model to dissect complex medical problems by deducing answers in a phased, in-depth manner. This mirrors the analytical approach of experienced clinicians, allowing the AI to handle multifaceted medical cases more effectively.

Achieving this level of sophistication necessitates high-quality reasoning training data. Quark has addressed this challenge by constructing a “dual data production line + dual incentive mechanism” engineering system, ensuring a consistent and reliable stream of pertinent medical information.

The Quark Health LLM is supported by a team of thousands of professionally trained medical annotators, with over 400 holding the credentials of Associate Chief Physician or higher. This commitment to expert oversight ensures the accuracy and reliability of the AI model’s outputs.

Leveraging its expertise, Quark’s AI search platform has attracted a substantial user base within the medical community.

As Zhao Cunzhong, head of Quark Health Operations, explains, “Our platform’s monthly active users among medical students nationwide has surpassed 2 million, representing over half of the total student population. They widely utilize Quark for foundational knowledge searches, exam preparation, and clinical diagnostic assistance.” This widespread adoption underscores the value of AI tools in supporting the next generation of healthcare professionals.