On Cognitive Maps, LLMs, World Models, and Understanding

In this hour-long talk from the Simons Institute's "The Future of Language Models and Transformers" series, Dileep George from Google DeepMind explores the relationships between cognitive maps, large language models (LLMs), world models, and understanding. Delve into how these concepts interconnect in artificial intelligence research and what implications they have for developing systems with deeper comprehension capabilities. Learn about cutting-edge perspectives on how AI systems might develop representations similar to the cognitive maps found in biological intelligence.