Logical forms complement probability in understanding language model (and human) performance

Yixuan Wang, Freda Shi

With the increasing interest in using large language models (LLMs) for planning in natural language, understanding their behaviors becomes an important research question. This work…