BREAKING: Apple just proved AI "reasoning" models like Claude, DeepSeek-R1, and o3-mini don't actually reason at all.
They just memorize patterns really well.
Here's what Apple discovered:
(hint: we're not as close to AGI as the hype suggests) pic.twitter.com/9A4Lg5aQWO
— Ruben Hassid (@RubenHssd) June 7, 2025
Apple’s latest AI study has shaken the tech world, revealing that leading reasoning models like Claude, DeepSeek-R1, and OpenAI’s o3-mini do not actually reason at all. Instead, they rely on sophisticated pattern matching, which works well until problems become too complex.
The findings challenge the hype surrounding artificial general intelligence. Apple’s research, titled “The Illusion of Thinking,” tested AI models on controlled logic puzzles rather than traditional benchmarks. The results showed that all models collapsed when faced with high-complexity tasks, proving that their reasoning abilities are limited.
The study identified three distinct reasoning regimes. At low complexity, standard AI models performed better than those marketed for reasoning. At medium complexity, reasoning models showed improvement by explicitly laying out intermediate steps. However, at high complexity, every model failed catastrophically, with performance dropping to near zero.
Apple’s research exposed a fundamental flaw in reasoning models. When given a step-by-step algorithm to follow, the models still failed at the same complexity threshold. This means they could not even mimic explicit logical procedures reliably.
The implications for AI development are significant. The study suggests that today’s AI models are not truly reasoning but merely mimicking patterns. This aligns with concerns raised by AI experts, including Meta’s Chief AI Scientist Yann LeCun, who has warned that current AI architectures are fundamentally limited.
Sources:
https://www.simplymac.com/ai/apples-latest-ai-study-strikes-at-the-heart-of-reasoning-model-hype
https://ml-site.cdn-apple.com/papers/the-illusion-of-thinking.pdf