Spoke at AI Safety Forum 2024
Introduction: State of Technical AI Safety
State of the science: The International Scientific Report on the Safety of Advanced AI
The International Scientific Report on the Safety of Advanced AI describes the capabilities, risks, and technical approaches to address risks of increasingly capable general-purpose AI systems. In this session we will explore the best current scientific understanding of AI safety including the most important challenges and the most promising methods for making progress.
Perspectives on generalisation in the science of AI safety
Understanding how AI models learn and generalise is necessary for designing, building, and deploying AI safely. In this session, Daniel Murfet and Marcus Hutter explore mathematical frameworks that illuminate AI behaviour, with implications for technical approaches to AI safety.