The First Impression Problem: Internal Bias Triggers Overthinking in Reasoning Models

Researchers identified 'internal bias' as a key cause of overthinking in AI reasoning models, where models form preliminary guesses that conflict with systematic reasoning. The study found that excessive attention to input questions triggers redundant reasoning steps, and current mitigation methods have proven ineffective.

• →AI reasoning models exhibit overthinking behavior characterized by redundant reasoning steps that waste computation.
• →Internal bias occurs when models form immediate preliminary guesses about answers before systematic reasoning begins.
• →Counterfactual experiments showed that removing input questions after processing reduces redundant reasoning across complex tasks.
• →Excessive attention to input questions serves as the key mechanism through which internal bias influences reasoning trajectories.
• →Current methods aimed at mitigating overthinking have failed to eliminate the persistent influence of internal bias.