Generative Query Response Grounding With Source Traceability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Generative models often generate unreliable responses with errors leading to hallucinations.
Innovation Solution
An electronic device extracts candidate data related to a query, selects input data based on content, applies the query and selected data to a generative model, and displays a visual representation indicating the origin of the input data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If generative models are used to generate responses, then response generation capability is improved, but reliability deteriorates due to errors and hallucinations
Solution Approach 1:
The patent introduces an intermediary system between the query and the generative model that extracts candidate data from source data, selects appropriate input data, and provides visual representations of data origins. This intermediary layer ensures that the generative model receives verified input data while maintaining its response generation capability, thus improving reliability without sacrificing versatility.
Solution Approach 2:
The patent implements a feedback mechanism where the system provides visual representations indicating the origin of input data used in generated responses. This feedback loop allows users to verify the accuracy and source of information, enabling them to assess the reliability of responses and providing implicit feedback for improving future response generation.
2Loss of information
If visual representation of input data origin is added, then transparency is improved, but device complexity increases
Solution Approach 1:
The patent extracts only the essential information about data origins and presents it as simplified visual representations (such as icons or labels) next to generated responses. Instead of displaying complete source data or complex metadata, the system extracts key identifying information, thereby improving transparency while minimizing the increase in system complexity.
Solution Approach 2:
The patent employs visual indicators such as color-coded icons or labels to represent different data sources or confidence levels. This use of visual cues provides intuitive transparency about data origins without requiring complex interfaces or additional system components, thus improving information transparency with minimal complexity overhead.
Data Source
AI summary
An electronic device includes: a display; at least one processor, comprising processing circuitry; and a memory storing instructions, wherein at least one processor, individually and/or collectively, is configured to execute the instructions and to cause the electronic device to: based on obtaining a query for source data, extract a plurality of pieces of candidate data related to the query from the source data; select, from the extracted plurality of pieces of candidate data, input data based on contents of the plurality of pieces of candidate data; generate a response to the query by applying the query and the selected input data to a generative model; determine a partial response of the response derived from the selected input data of the response; and display, via the display, a visual representation indicating information regarding the selected input data in an area corresponding to the determined partial response.