LLM Text Generation Tool for Writer Pseudonymity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing authorship attribution technologies can reveal the identity of a writer through distinctive writing styles, posing a risk to the author's safety and privacy, especially for whistleblowers who need to remain pseudonymous.
Innovation Solution
A personal security tool using Large Language Models (LLMs) with a chatbot prompt stylometrics alert module, text compiling module, and objective editing module to generate text that conceals the author's features, ensuring pseudonymity by preventing the LLM from assimilating personal features and allowing only objective edits.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a writer uses distinctive writing styles to express personal voice, then the writing quality and authenticity are improved, but the author's identity can be revealed through authorship attribution technologies
Solution Approach 1:
The patent introduces an intermediary system consisting of multiple LLMs that act as mediators between the author and the final text. The author's ideas are transformed through multiple LLM processors that collectively erase individual writing style markers, producing text that cannot be attributed to any single author while still conveying the original message
Solution Approach 2:
The writing process is segmented into multiple independent stages, each handled by a different LLM. The text is broken down and reprocessed through multiple models (e.g., ChatGPT, Bard, Gemini) to distribute and dilute any single author's stylistic fingerprint across the final output
2Object-affected harmful factors
If an LLM is used to generate text to conceal authorship features, then pseudonymity is preserved, but the LLM may assimilate the user's features through interaction
Solution Approach 1:
The system performs preliminary actions by pre-processing the author's input ideas through multiple LLMs before the final text generation. This preliminary transformation through multiple models prevents any single LLM from assimilating the user's features, as the input to each model has already been modified by previous models
Solution Approach 2:
Different LLMs are assigned to different local processing stages of the text generation pipeline. Each model processes specific portions or aspects of the content, preventing any single model from absorbing the complete set of user features while still contributing to the overall pseudonymous output
3Object-affected harmful factors
If multiple LLMs are used to process text to enhance pseudonymity, then authorship attribution resistance is improved, but the device complexity increases
Solution Approach 1:
The patent employs multiple LLMs that serve universal functions in the text processing pipeline. Each model can perform multiple roles (style transformation, content generation, refinement) depending on the processing stage, reducing the need for specialized components and simplifying the overall system architecture despite using multiple models
Data Source
AI summary
The present invention is a personal security method and a writer's tool, in two or three modules, that assures generation of text by a user which cannot be identified has having been written by said user. The present invention uses Large Language Modules (LLMs such as but not limited to ChatGPT) principally to generated the document. Using an LLM together with the two or three modules of the present invention side-steps the author's (user's) features entirely by having the LLM generate the text, creating an authorship opaque document, and thus concealing and securing the identity of the writer.