overhyping large language models right now, not seeing the real value yet