LLM Prompt Data Source Curation for Privacy-Compliant Summaries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Large language models (LLMs) face challenges in generating summaries that comply with privacy rules and personal preferences, especially in multi-jurisdictional systems where legal and personal preferences vary, leading to governance complexities.
Innovation Solution
A user interface (UI) is used to present data source identifications and selections, allowing users to curate data sources based on legal and personal preferences, with real-time approval and persistence of selections, ensuring compliance and privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If LLM generates summaries using multiple data sources to improve comprehensiveness, then information completeness is improved, but privacy compliance and governance complexity worsen
Solution Approach 1:
The patent segments the data source selection process into individual controllable units. Each data source is presented as a separate identification with its own selection indicator, allowing granular control over which data sources are used. This segmentation enables comprehensive information gathering while maintaining simplified governance through user-directed choices rather than complex system-wide policies.
Solution Approach 2:
The patent implements preliminary action by obtaining user approvals and curating data source selections before the LLM generation process. The system presents data source identifications and selections to the user in advance, allowing them to approve or reject specific sources based on privacy preferences and legal requirements. This pre-curation simplifies governance by establishing clear boundaries before information processing begins.
2Manufacturing precision
If LLM uses all available data sources to improve summary quality, then output quality is improved, but processing load and bandwidth consumption worsen
Solution Approach 1:
The patent extracts and removes unnecessary data sources from the processing pipeline through user-based curation. By allowing users to select only the data sources they want used, the system eliminates redundant or unwanted data from being processed by the LLM. This extraction maintains summary quality by keeping relevant sources while reducing processing load by excluding irrelevant ones.
Solution Approach 2:
The patent applies partial action by using only the subset of data sources that users explicitly select and approve, rather than processing all available data sources. This partial approach to data usage maintains sufficient summary quality for user needs while significantly reducing bandwidth consumption and processing load by limiting the scope of data transformation and model inference.
3Adaptability or versatility
If LLM generates personalized summaries according to individual preferences, then user satisfaction is improved, but system complexity and governance burden worsen
Solution Approach 1:
The patent implements self-service by enabling users to directly control which data sources are used for their personalized summaries. The system presents data source identifications and selection indicators to users, who then make their own choices based on personal preferences and privacy concerns. This self-service approach achieves high personalization capability without increasing system complexity, as the automation burden shifts to the user interface rather than the backend governance infrastructure.
Data Source
AI summary
Data source curation for large language model (LLM, generically multimodal model) prompts enables compliance with privacy requirements, while allowing for personal preferences, when Generated passages relating to a person import information from external data sources. A set of data sources is presented to a user, permitting the user to select which to use or avoid using. In some examples, when a user adds a data source, based on the nature of the data source, the user is provided a prompt to provide consent for the use of the content of that data source, enabling user-directed curation of the data sources used in the LLM generation of a summary about the person. In some examples, the set of data sources presented to a user had been previously curated according to governing policies, to restrict the user from accepting use of a data source that is disfavored for legal or policy reasons.