The research team from Peking University, in collaboration with researchers from multiple universities both domestically and internationally, has released a comprehensive survey on AI alignment, covering four core issues for achieving AI alignment: “Learning from Feedback,” “Learning under Distributional Shift,” “Assurance,” and “AI Governance.” The survey proposes that AI alignment is a continuously updating, iteratively improving loop.
Yawen Duan, Kwan Yee Ng, and Brian Tse from Concordia AI contributed to the overall direction of the survey, the content framework, and the AI governance section in Chapter 5.