Understanding and Steering Generative AI Systems

Watch a lecture from UC Berkeley's Jacob Steinhardt exploring the societal impact and technical challenges of generative AI systems, with a focus on developing tools for understanding and controlling their behavior. Learn about the rapid growth and adoption of large language models and vision-language models, while examining methods to systematically identify unexpected behaviors and categorize them into interpretable patterns. Discover approaches for leveraging AI systems to analyze other AI models, and explore techniques for improving model accuracy and truthfulness by understanding neural representations. Gain insights into the complexities of open-ended AI behavior and the importance of developing robust tools for societal oversight of these increasingly prevalent technologies.