Persuasion Tokens for Editing Factual Knowledge in LLMs
This provides a more practical and scalable method for editing factual knowledge in LLMs, addressing a key bottleneck for researchers and practitioners.
The paper tackles the inefficiency of in-context knowledge editing (IKE) in LLMs by introducing persuasion tokens (P-Tokens) that replicate IKE effects without fact-specific demonstrations, achieving comparable or better performance across datasets and models.
In-context knowledge editing (IKE) is a promising technique for updating Large Language Models (LLMs) with new information. However, IKE relies on lengthy, fact-specific demonstrations which are costly to create and consume significant context window space. In this paper, we introduce persuasion tokens (P-Tokens) -- special tokens trained to replicate the effect of IKE demonstrations, enabling efficient knowledge editing without requiring fact-specific demonstrations. We evaluate P-Tokens across two editing datasets and three LLMs, demonstrating performance comparable to, and often exceeding, IKE. We further find that editing performance is robust to distractors with small negative effects to neighboring facts, and that increasing the number of P-Tokens improves performance. Our work addresses key limitations of IKE and provides a more practical and scalable alternative for editing LLMs.