Shanghai, March 16, 2026 – The Xuemin Institute of Advanced Studies at Fudan University hosted the "Xuemin Lecture," featuring Prof. Stephen Hsu, a distinguished theoretical physicist and entrepreneur from Michigan State University. The event, moderated by Executive Dean Prof. Xiaohong Yang, focused on the transformative impact of generative AI on basic scientific research.

Prof. Hsu, whose career spans academia and multiple successful startups, presented a compelling case for AI as a true research collaborator rather than a mere tool. He highlighted a striking example: while top-tier general models like Gemini and GPT struggled with complex International Mathematical Olympiad problems, an "Agent Harness"—a generate-and-verify system—boosted their success rate from 1/6 to 5/6. This iterative process of generation and verification drastically reduces error rates, making AI a reliable partner in scientific discovery.

Drawing from his own research in quantum mechanics, Prof. Hsu demonstrated how AI can provide breakthrough insights across disciplines, accelerating the research process. "It's like having a brilliant but somewhat unreliable colleague," he noted, emphasizing the need to embrace AI's potential despite its imperfections.

However, Prof. Hsu identified a critical missing element in current AI models: "Scientific Taste." This advanced judgment, cultivated over decades by seasoned researchers, is essential for identifying valuable research directions and spotting hidden issues. He revealed ongoing collaborations with top AI labs to instill this "taste" into models by capturing interactions between scientists and AI through tools like LaTeX editors.

The lecture also addressed the broader implications of AI on the academic ecosystem. Prof. Hsu warned of the potential for "AI junk papers" from those lacking domain expertise, while also noting a shift in mentorship models as senior researchers increasingly collaborate with AI over guiding doctoral students.

During the Q&A, Prof. Hsu discussed the future of AI in academia, suggesting that acknowledging AI's contributions in publications will become standard practice. He concluded with a visionary outlook: "We will eventually reach a point where different AI agents autonomously discuss the most cutting-edge scientific problems... We are undoubtedly on this path of transformation."

The Xuemin Lecture continues to foster the exchange of cutting-edge ideas, aligning with the Institute's mission to support innovative young scholars in making revolutionary discoveries.