Reasoning Abilities of Large Language Models: In-Depth Analysis on the Abstraction and Reasoning Corpus