🚀 Thrilled to announce I'll be joining NYU CS Department as first PhD student of Prof. Sherry Yang! Honored to be pioneering her lab's research in world models, working on general intelligence in both language and robotics. Stay tuned for what we build! #NYU #PhDLife

I am also presenting this work at #NeurIPS2024! Find me at:
Future NLP Workshop: Dec 10, 11:25-12:25, Theatre C300, UBC Robson Square
Compositional Learning Workshop: Dec 15, Room 118-120
Interested in tree search for llm, planning w/ world models, or test time scale? Feel free to contact!

We characterize this CoT approach as 'Cognitive Maps for Path Planning' since it mirrors key aspects of human cognition, including extrapolation and rapid adaptation. We expect the development of a generalized cognitive map for language models to be a promising next step

Tl;dr:
Standard planning methods, whether implicit or with conventional CoT, fail to extrapolate beyond training data. However, we found a CoT variant configured with sampling, propagation, and backtracking, enabling robust planning with world model in novel environments.

Humans excel at extrapolating from simple input—far beyond a ‘stochastic parrot.’ Are language models destined to fall short?

This year, I explored this question primarily in the Gridworld domain. Wrapping up my research, I made a blog including key results and takeaways:
bit.ly/3ZpJQUw