> Apple's recent research paper, The Illusion of Thinking, tested modern LLMs on classic logic puzzles—Tower of Hanoi, river-crossing tasks, and other structured reasoning challenges.
> The results? Performance dropped to zero.
> Children can solve these puzzles. AI could not.
Ah, no.
The AI that Apple tested in this paper did drop to zero, but on tasks where the output would exceed the length of the output context window (Tower of Hanoi when the towers are tall enough), or where the answer was "this is provably impossible" (some of the river-crossing tasks), as per the abstract in this response: https://arxiv.org/abs/2506.09250
ben_w•7mo ago
> Apple's recent research paper, The Illusion of Thinking, tested modern LLMs on classic logic puzzles—Tower of Hanoi, river-crossing tasks, and other structured reasoning challenges.
> The results? Performance dropped to zero.
> Children can solve these puzzles. AI could not.
Ah, no.
The AI that Apple tested in this paper did drop to zero, but on tasks where the output would exceed the length of the output context window (Tower of Hanoi when the towers are tall enough), or where the answer was "this is provably impossible" (some of the river-crossing tasks), as per the abstract in this response: https://arxiv.org/abs/2506.09250