Peking University's Yuan Li on What Universities Can Research in the AI Boom
In brief
Yuan Li of Peking University argues that universities need not chase leading companies in compute and data; they can explore new architectures around real problems and advance AI research through open source and industry collaboration. Open-Sora Plan emphasizes opening its architecture, code and training methods and documenting trial-and-error experience, not replicating Sora's final results; Helios explores streaming video generation, and OpenAI4S lets code run in persistent Python or R environments while keeping intermediate objects and version records.
Why it matters
Interview shows how a university team with limited compute advances video generation and scientific agents by open-sourcing architecture, code and training methods.
