1 paper · 1 filter
Hoang Anh Just, Mahavir Dabas, Lifu Huang +2
Existing work on improving language model reasoning typically explores a single solution path, which can be prone to errors. Inspired by perspective-taking in social studies, this…