{"ID":6267172,"CreatedAt":"2026-07-10T01:11:38.759438437Z","UpdatedAt":"2026-07-14T02:54:43.516908796Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2607.08393","arxiv_id":"2607.08393","title":"Towards Mechanistically Understanding Why Memorized Knowledge Fails to Generalize in Large Language Model Finetuning","abstract":"Fine-tuning LLMs to inject new knowledge faces a critical challenge: LLMs can quickly memorize new facts, yet fail to use them for downstream reasoning tasks. We formalize this failure as the \\textit{\\textbf{Knowing--Using Gap}}, characterized by an accuracy gap and a temporal lag between memorization and generalization. To understand this phenomenon, we fine-tune LLMs with unseen knowledge and monitor the spatial permeation dynamics of the knowledge internally using a novel intervention technique called self-patching. Self-patching identifies activation locations where relocating representations substantially improves failed generalization cases. These results are consistent with a knowledge-circuit misalignment hypothesis: memorized representations can exist internally but may not be routed to computation-effective layers. To demonstrate the practicality of this diagnostic finding, we design a simple heuristic strategy which recovers 58--75\\% of the oracle headroom in generalization failure. Experiments are done cross-domain for the robustness of this finding.","short_abstract":"Fine-tuning LLMs to inject new knowledge faces a critical challenge: LLMs can quickly memorize new facts, yet fail to use them for downstream reasoning tasks. We formalize this failure as the \\textit{\\textbf{Knowing--Using Gap}}, characterized by an accuracy gap and a temporal lag between memorization and generalizatio...","url_abs":"https://arxiv.org/abs/2607.08393","url_pdf":"https://arxiv.org/pdf/2607.08393v1","authors":"[\"Lu Dai\",\"Ziyang Rao\",\"Yili Wang\",\"Hanqing Wang\",\"Hao Liu\",\"Hui Xiong\"]","published":"2026-07-09T12:17:28Z","proceeding":"cs.AI","tasks":"[\"cs.AI\",\"cs.CL\"]","methods":"[\"Large Language Model\",\"Language Model\"]","has_code":false}