“Current state-of-the-art LLMs possess reasoning capabilities but lack the crispness and accuracy to avoid hallucinations or consistently solve basic puzzles, even when they can solve complex problems.”