Open-Source and Science in the Era of Foundation Models

Watch a thought-provoking lecture from Stanford University's Percy Liang exploring the critical intersection of open-source development and foundation models in AI research. Examine how increasing model capabilities often correlate with decreased openness, and discover why open-source models are fundamental for establishing a rigorous AI foundation. Learn about the different levels of model access - from API to open-weight to open-source - and their unique contributions to research capabilities. Explore recent developments in simulation and problem-solving agents through API access, understand how open weights enable safety and interpretability research through "model forensics," and see how open-source access drives innovation in architectures, training procedures, and data curation. Address the key challenges of building open-source models, including resource requirements in data, compute power, and talent, while discovering promising community-driven approaches that could help realize the vision of open-source foundation models.