Probabilistic Thinking in Language and Code

Watch a 50-minute research lecture exploring the intersection of Bayesian cognitive models and Large Language Models (LLMs), delivered at UCLA's Institute for Pure & Applied Mathematics. Discover novel approaches to bridging probabilistic reasoning with both natural and programming languages as potential languages-of-thought for human-like representations. Examine a specialized class of Bayesian models integrated with LLMs, understanding how these hybrid systems exhibit more human-like characteristics compared to standalone LLMs or traditional Bayesian cognitive models. Learn about wake-sleep learning techniques for fine-tuning language models to enhance their inductive reasoning capabilities through probabilistic inference amortization. Presented by Cornell University researcher Kevin Ellis at the Naturalistic Approaches to Artificial Intelligence Workshop.