The Curse of Popularity: Popular Entities have Catastrophic Side Effects when Deleting Knowledge from Language Models
This addresses privacy concerns in AI by revealing a critical issue in knowledge deletion, though it is incremental as it builds on existing deletion research.
The study tackled the problem of catastrophic side effects when deleting knowledge from language models, finding that deleting knowledge related to popular entities leads to severe side effects, with analysis conducted on models trained on synthetic knowledge graphs.
Language models (LMs) encode world knowledge in their internal parameters through training. However, LMs may learn personal and confidential information from the training data, leading to privacy concerns such as data leakage. Therefore, research on knowledge deletion from LMs is essential. This study focuses on the knowledge stored in LMs and analyzes the relationship between the side effects of knowledge deletion and the entities related to the knowledge. Our findings reveal that deleting knowledge related to popular entities can have catastrophic side effects. Furthermore, this research is the first to analyze knowledge deletion in models trained on synthetic knowledge graphs, indicating a new direction for controlled experiments.