They basically tried to make LRMs solve puzzles after teaching them the algorithm, which they failed.

For those who want to: readhttps://machinelearning.apple.com/research/illusion-of-thinking