Measurements for Capabilities and Hazards

This talk by Dan Hendrycks from the Center for AI Safety explores methodologies for measuring both the capabilities and potential hazards of AI systems. Delivered at the Simons Institute as part of the Safety-Guaranteed LLMs series, the presentation delves into quantitative frameworks for evaluating AI performance alongside associated risks. Learn about assessment techniques that help researchers and developers understand not only what AI systems can accomplish but also where they might pose dangers, providing crucial insights for building safer artificial intelligence technologies.