Stanford CS25: V5 I RL as a Co-Design of Product and Research, Karina Nguyen

Karina Nguyen from OpenAI delivers a lecture on the co-design of product and research in AI, highlighting AI's potential in education, personalized tool creation, and augmented creativity. She discusses two scaling paradigms: next token prediction and chain of thought, emphasizing the importance of building and creating with AI. Nguyen shares her experience at Anthropic and OpenAI, explaining two approaches to building research-driven products: creating familiar form factors for unfamiliar capabilities and training models to fulfill a specific vision. She details a case study on shaping model behavior around refusals, including crafting charitable interpretations and using non-violent communication. The lecture concludes with a Q&A session, addressing topics such as inducing features into models, data collection, reward design, and the future of human-AI interactions, including dynamic generative UI and personalized access to healthcare and education.