Sendhil Mullainathan

What's Producible May Not Be Reachable: Measuring the Steerability of Generative Models

What Has a Foundation Model Found? Using Inductive Bias to Probe for World Models

Potemkin Understanding in Large Language Models

Evaluating the World Models Used by Pretrained Learners

Language Generation in the Limit

Evaluating the World Model Implicit in a Generative Model