Answer-Centric or Reasoning-Driven? Uncovering the Latent Memory Anchor in LLMs
This work challenges the perceived reasoning depth of LLMs, which is crucial for AI researchers and developers seeking to improve model interpretability and reliability.
The study investigated whether large language models rely on memorized answer patterns rather than genuine inference, finding that performance dropped by 26.90% when answer cues were masked, indicating post-hoc rationalization.
While Large Language Models (LLMs) demonstrate impressive reasoning capabilities, growing evidence suggests much of their success stems from memorized answer-reasoning patterns rather than genuine inference. In this work, we investigate a central question: are LLMs primarily anchored to final answers or to the textual pattern of reasoning chains? We propose a five-level answer-visibility prompt framework that systematically manipulates answer cues and probes model behavior through indirect, behavioral analysis. Experiments across state-of-the-art LLMs reveal a strong and consistent reliance on explicit answers. The performance drops by 26.90\% when answer cues are masked, even with complete reasoning chains. These findings suggest that much of the reasoning exhibited by LLMs may reflect post-hoc rationalization rather than true inference, calling into question their inferential depth. Our study uncovers the answer-anchoring phenomenon with rigorous empirical validation and underscores the need for a more nuanced understanding of what constitutes reasoning in LLMs.