<ul data-eligibleForWebStory="true"><li>Apple researchers published a study on simulated reasoning models aligning responses with training data.</li><li>Findings similar to USAMO study where AI models scored low on novel mathematical proofs.</li><li>The study titled 'The Illusion of Thinking' by an Apple team examined 'large reasoning models.'</li><li>Models attempted logical reasoning to solve problems step-by-step through text output.</li><li>AI models were tested on classic puzzles of varying complexity including Tower of Hanoi and checkers jumping.</li><li>Evaluation critique highlights a focus on final answer accuracy rather than the reasoning process.</li></ul>

New Apple study challenges whether AI models truly think through problems

Discover more