Prompt information enhancement method and related device

CN117332040BActive Publication Date: 2026-07-03BEIJING QIHOOD TECHNOLOGY CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
BEIJING QIHOOD TECHNOLOGY CO LTD
Filing Date
2023-09-27
Publication Date
2026-07-03

AI Technical Summary

Technical Problem

Large language models cannot accurately distinguish between task attribute information and task execution information in practical applications. This allows attackers to bypass security check mechanisms by injecting malicious instructions, generating malicious content or non-compliant responses, which poses a potential threat to users and applications.

Method used

By performing multiple enhancement processing on the prompts, multiple enhanced prompts are generated. The target enhanced prompt is selected based on preset rules to form a prompt enhancement strategy, thereby improving the security performance of large language models.

Benefits of technology

It improves the protection performance of large language models in different scenarios, enhances the accuracy of understanding and executing task attribute information, and meets security protection requirements.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117332040B_ABST
    Figure CN117332040B_ABST
Patent Text Reader

Abstract

This application discloses a method for generating a prompt enhancement strategy, a method for enhancing prompt information, and related equipment, relating to a natural language processing module. The method includes: acquiring prompt information including task attribute information and task execution information; performing multiple prompt enhancement processes on the prompt information to obtain multiple enhanced prompt information corresponding to the prompt information; selecting at least one target enhanced prompt information from the multiple enhanced prompt information based on preset rules to obtain a prompt enhancement strategy including the at least one target enhanced prompt information; wherein the prompt enhancement strategy is used to instruct the large language model to execute the task corresponding to the task attribute information based on the task execution information. This application proposes a method for dynamically selecting multiple enhanced prompt information to obtain a prompt enhancement strategy, improving the rationality and accuracy of selecting the prompt enhancement strategy and meeting the security protection requirements of large language models.
Need to check novelty before this filing date? Find Prior Art