IC Colloquium: Testing AI's Implicit World Models
Par : Keyon Vafa - Harvard University
IC Faculty candidate
Abstract
Real-world AI systems must be robust across a wide range of conditions. One path to such robustness is if a model recovers a coherent structural understanding of its domain. But it is unclear how to measure, or even define, structural understanding. This talk will present theoretically-grounded definitions and metrics that test the structural recovery — or implicit “world models” — of generative models. We will propose different ways to formalize the concept of a world model, develop tests based on these notions, and apply them across domains. In applications ranging from testing whether LLMs apply logic to whether foundation models acquire Newtonian mechanics, we will see that models can make highly accurate predictions with incoherent world models. We will also connect these tests to a broader agenda of building generative models that are robust across downstream uses, incorporating ideas from statistics and the behavioral sciences. Developing reliable inferences about model behavior across tasks offer new ways to assess and improve the efficacy of generative models.
Bio
Keyon Vafa is a postdoctoral fellow at Harvard University and an affiliate with the Laboratory for Information & Decision Systems at MIT. His research focuses on understanding and improving the implicit world models learned by generative models. Keyon completed his PhD in computer science from Columbia University, where he was an NSF GRFP Fellow and the recipient of the Morton B. Friedman Memorial Prize for excellence in engineering. He has organized the NeurIPS 2024 Workshop on Behavioral Machine Learning and the ICML 2025 Workshop on Assessing World Models, and serves on the Early Career Board of the Harvard Data Science Review.
More information
IC Faculty candidate
Abstract
Real-world AI systems must be robust across a wide range of conditions. One path to such robustness is if a model recovers a coherent structural understanding of its domain. But it is unclear how to measure, or even define, structural understanding. This talk will present theoretically-grounded definitions and metrics that test the structural recovery — or implicit “world models” — of generative models. We will propose different ways to formalize the concept of a world model, develop tests based on these notions, and apply them across domains. In applications ranging from testing whether LLMs apply logic to whether foundation models acquire Newtonian mechanics, we will see that models can make highly accurate predictions with incoherent world models. We will also connect these tests to a broader agenda of building generative models that are robust across downstream uses, incorporating ideas from statistics and the behavioral sciences. Developing reliable inferences about model behavior across tasks offer new ways to assess and improve the efficacy of generative models.
Bio
Keyon Vafa is a postdoctoral fellow at Harvard University and an affiliate with the Laboratory for Information & Decision Systems at MIT. His research focuses on understanding and improving the implicit world models learned by generative models. Keyon completed his PhD in computer science from Columbia University, where he was an NSF GRFP Fellow and the recipient of the Morton B. Friedman Memorial Prize for excellence in engineering. He has organized the NeurIPS 2024 Workshop on Behavioral Machine Learning and the ICML 2025 Workshop on Assessing World Models, and serves on the Early Career Board of the Harvard Data Science Review.
More information
Practical information
- General public
- Free
Contact
- Host: TBD