Generalizable Long-Horizon Manipulations with Large Language Models