<ul><li>Latest large reasoning models (LRMs) face 'complete accuracy collapse' on highly complex tasks, as per a new paper co-authored by Apple researchers.</li><li>LRMs perform well on some problems but struggle on highly complicated ones, indicating a limit to their performance.</li><li>Researchers used controllable puzzles to analyze LRMs' reasoning breakdown as difficulty increased, leading to 'complete accuracy collapse' beyond a certain complexity level.</li><li>Study reveals LRMs reduce reasoning effort on harder problems, showcasing fundamental limitations in how they handle complex tasks, raising questions about achieving truly general AI.</li></ul>

Report: Reasoning AI Models Experience ‘Complete Accuracy Collapse’ When Problems Get Too Complicated

Discover more