Persuasion Tokens for Editing Factual Knowledge in LLMs

2026-01-23

AI Summary

This work proposes “persuasion tokens” (P-Tokens)—learnable, universal tokens that replace fact-specific examples in in-context knowledge editing (IKE). By leveraging end-to-end training, P-Tokens enable efficient knowledge injection without consuming additional context length, addressing the high cost and excessive context window usage of conventional IKE methods. The approach supports compositional optimization through multiple P-Tokens and achieves performance on par with or superior to IKE across two knowledge editing benchmarks and three large language models. Furthermore, it demonstrates robustness against distractors and exhibits scalable improvements as the number of P-Tokens increases, significantly enhancing both the efficiency and scalability of knowledge editing.

Technology Category

• Natural Language Processing: Prompt Engineering / Prompting
• Search and Optimization: Learning to Search
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

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.

Problem

• in-context knowledge editing
• Large Language Models
• factual knowledge editing
• context window efficiency
• knowledge updating

Innovation

• Persuasion Tokens
• In-context Knowledge Editing
• Large Language Models
• Knowledge Editing
• Efficient Prompting