Can AI Think? The "Human-Like" Claim About the Centaur Model Comes Under Fire
A Zhejiang University study finds that the Centaur model's 160-task performance likely stems from overfitting; the "pick option A" test exposes pattern matching, not understanding.