arxiv preprint - The Truth is in There: Improving Reasoning in Language Models with Layer-Selective Rank Reduction | AI Breakdown | Podwise