You could achieve thinking though logical problems by adding chain of thought or tree of thought. Then the LLM will try to break stuff down into logic it’s already seen.