<ul><li>Apple researchers found 'fundamental limitations' in cutting-edge AI models, with large reasoning models facing a 'complete accuracy collapse' in highly complex problems.</li><li>Standard AI models performed better than LRMs in low-complexity tasks, while both suffered 'complete collapse' with high-complexity tasks.</li><li>LRMs reduce reasoning effort as they near performance collapse, leading to concerns raised by Apple researchers and AI experts like Gary Marcus.</li><li>The study indicates a 'fundamental scaling limitation' in the thinking capabilities of current reasoning models, with implications for the development of AI towards artificial general intelligence (AGI).</li></ul>

Advanced AI suffers ‘complete accuracy collapse’ in face of complex problems, study finds

Discover more