Can Fine-Tuning Erase Your Edits? On the Fragile Coexistence of Knowledge Editing and Adaptation
arXiv:2511.05852v4 Announce Type: replace-cross Abstract: Knowledge editing (KE) offers a lightweight alternative to retraining for updating large language models (LLMs). Meanwhile, fine-tuning remains the default operation for adapting LLMs to new domains and tasks. Despite their widespread adoption, these two post-training interventions have been studied in isolation, leaving open a crucial question: if we fine-tune an edited model, do the edits survive? This question is motivated by practical objectives: removing covert or malicious edits, and preserving beneficial edits. If fine-tuning imp
This research addresses a critical and previously unexplored interaction between two fundamental LLM post-training interventions as AI models become more complex and widely deployed.
Understanding the fragility of knowledge edits during fine-tuning is crucial for maintaining model integrity, preventing unintended behaviors, and ensuring the reliability of AI systems in sensitive applications.
This research highlights that knowledge edits are not persistently embedded and can be erased, necessitating more robust editing and adaptation strategies for LLMs.
- · Developers of robust knowledge editing techniques
- · Organizations relying on precise LLM behavior control
- · AI safety and alignment researchers
- · Developers using simplistic knowledge editing methods
- · Users unaware of edit fragility
Increased research into resilient knowledge editing techniques and improved methods for tracking model provenance.
Development of new LLM architectures or training methodologies that better preserve critical knowledge during adaptive fine-tuning.
Potential for new regulatory frameworks or standards for LLM maintainability and predictable behavior in regulated industries.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.AI