Chris Maddison

The success of large language models is driven by the abundance and natural structure of text. What does this tell us about our universe and ourselves? How can we use these insights to advance applications in other domains when data is scarce? I want to understand how the statistical structure of real-world data influences the emergence of capabilities in AIs as they train on vast, heterogeneous datasets. I also want to develop data-efficient learning systems and, with them, build foundational AIs for medicine and biology. I tend to publish at machine learning conferences (NeurIPS, ICML, ICLR).