| title | Towards Mechanistically Understanding Why Memorized Knowledge Fails to Generalize in Large Language Model Finetuning |
|---|---|
| authors | Lu Dai, Ziyang Rao, Yili Wang, Hanqing Wang, Hao Liu, Hui Xiong |
| date | 2026-07-09 |
| venue | arXiv (preprint, not peer reviewed) |
| id | arXiv:2607.08393 |
| affiliation | HKUST |
| upvotes | 13 (Hugging Face Daily Papers, 2026-11-01) |
| links | abs pdf hf |
| 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 \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. |
cite
@misc{dai2026towards,
title={Towards Mechanistically Understanding Why Memorized Knowledge Fails to Generalize in Large Language Model Finetuning},
author={Lu Dai and Ziyang Rao and Yili Wang and Hanqing Wang and Hao Liu and Hui Xiong},
year={2026},
eprint={2607.08393},
archivePrefix={arXiv},
url={https://arxiv.org/abs/2607.08393}
}