David Jensen
London, E1 8PH, UK
Portland, ME 04101
2nd floor
11th floor
Boston, MA 02115
2nd floor
Talk recording
Large language models (LLMs) and other foundation models have enabled a set of unexpected and socially transformative AI technologies. This has elevated concerns about the potential near-term impacts of AI and produced calls for immediate regulation of these technologies. However, our ability to forecast the impacts of AI and craft appropriate responses depends critically on the underlying science — what is known about fundamental computational mechanisms that enable model capabilities, how AI systems will respond to varying inputs and contexts, and how these systems can be guided and controlled. Unfortunately, scientific research in modern AI is hindered by a set of scientific, institutional, and economic challenges. Taken together, these challenges present barriers to understanding that are unprecedented in the history of modern science.



