Can multiple-choice questions really be useful in detecting the abilities of LLMs?