New Apple study challenges whether AI models truly “reason” through problems - Ars Technica

Puzzle-based experiments reveal limitations of simulated reasoning, but others dispute findings.