Percy Liang on the Center for Research on Foundation Models

Foundation models, such as GPT-3, represent a paradigm shift in AI, moving from specialized classification tasks to general-purpose, emergent capabilities. The Center for Research on Foundation Models (CRFM) at Stanford addresses the challenges posed by these models, emphasizing the need for interdisciplinary academic oversight to ensure safety, transparency, and societal benefit. While scaling compute and data remains a primary driver of model performance, technical progress must prioritize explainability and robustness against adversarial exploits. Academia serves a critical role as an independent auditor, establishing community norms and benchmarking standards that go beyond commercial objectives. Future development requires a holistic approach, integrating human-in-the-loop systems and rigorous evaluation protocols to mitigate risks like bias and misinformation, while navigating the complex ethical landscape of data usage and ownership in an increasingly automated world.