Together Talks | Ep 1: Percy Liang on Open Source, AI Evaluations, and the Future of LLMs

The evolution of AI research has shifted from a transparent, collaborative environment in the 2010s to a fragmented, closed-off industry dominated by proprietary models. Dr. Percy Liang, a professor at Stanford and director of the Center for Research on Foundation Models, argues that the future of AI requires a "Linux for foundation models"—a decentralized, fully transparent ecosystem where training data and code are accessible. This openness is essential not only for innovation but for rigorous safety and auditability, as current "security by obscurity" practices often mask risks. Effective progress depends on objective evaluation frameworks, such as the Holistic Evaluation of Language Models, to address challenges like data contamination and the difficulty of measuring true model intelligence. Establishing a sustainable, community-driven approach to funding and data creation is critical to maintaining an open frontier in AI development.