Apple: "The Illusion of Thinking"

Apple researchers report reasoning models collapse to zero accuracy beyond a complexity threshold on controllable puzzles.

What happened and why it matters

The paper argued that reasoning models don't develop generalisable problem-solving. It was heavily disputed — critics showed failures were partly due to output-length limits and impossible puzzle instances — so it is scored as a modest negative.

Sources

• The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity ↗ Apple Machine Learning Research · 7 Jun 2025