Prompt Creation Device for Jailbreak Detection via Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Large language models (LLMs) face challenges in generating inappropriate answers, including unethical or dangerous content, when encountering jailbreak prompts, and existing solutions fail to effectively address these issues.
Innovation Solution
A prompt creation device and response system that analyze input prompts using natural language processing to break them down into simpler structures, generating reconstructed prompts that focus on the instruction and data, thereby improving the likelihood of rejecting inappropriate answers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the LLM processes complex input prompts with background and context information, then the answer accuracy is improved, but the risk of generating inappropriate answers through jailbreak prompts increases
Solution Approach 1:
The input prompt is segmented into instruction and background components. The system creates a simplified prompt containing only the instruction, processes it through the LLM, then compares the result with the original complex prompt to detect jailbreak attempts.
Solution Approach 2:
A prompt creation device acts as an intermediary between the user input and the LLM. It processes the input prompt to create a reconstructed prompt that removes potentially harmful background information while preserving the core instruction, thereby mediating the interaction to prevent jailbreak attacks.
2Object-affected harmful factors
If the system removes background and context information from prompts, then the risk of jailbreak prompts is reduced, but the answer accuracy deteriorates
Solution Approach 1:
The system performs preliminary processing by creating a simplified prompt from the input prompt before submitting it to the LLM. This preliminary action removes potentially harmful background information while preserving the essential instruction, preventing jailbreak attacks before they can affect the LLM.
Solution Approach 2:
The system changes the parameters of the prompt by selectively removing background and context information while retaining the core instruction. This parameter modification transforms the complex input prompt into a simplified version that is safer for LLM processing while maintaining answerability for legitimate queries.
3Reliability
If the system processes and reconstructs prompts before LLM input, then the probability of rejecting inappropriate answers is improved, but the processing complexity increases
Solution Approach 1:
The system extracts only the essential instruction component from the complex input prompt, separating it from the background and context information. This extraction process simplifies the prompt while maintaining the core query, reducing processing complexity compared to analyzing and evaluating the entire complex prompt structure.
Data Source
AI summary
A prompt creation device creating a requesting prompt, from an input prompt including an instruction and at least one of background and input data, the requesting prompt requesting an answer only to the instruction and the input data indicating the content of the instruction.