Language Model Source Disclosure for Response Reliability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Users are unable to understand the reliability of information provided by conventional systems due to a lack of transparency regarding the sources used for generating responses.
Innovation Solution
An information providing apparatus that utilizes multiple language models trained on different data sources, providing responses with source information and differentiated service plans to enhance user understanding of reliability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a generative model is used to generate response text, then the system can provide automated responses to user questions, but the user cannot understand the reliability of the information because the source is not disclosed
Solution Approach 1:
The system segments the information provision process into two distinct components: the generative model that creates responses and the separate source information that documents data origins. This segmentation allows the response generation to remain automated while the source information is separately tracked and presented to users, resolving the contradiction between automation and transparency.
Solution Approach 2:
The patent introduces an intermediary mechanism (the source information provision unit) that acts as a mediator between the generative model and the user. This intermediary captures, stores, and presents source information without interfering with the automated response generation process, enabling users to understand information reliability while maintaining system productivity.
2Reliability
If multiple language models with different data sources are used, then users can assess information reliability through source transparency, but the system complexity increases
Solution Approach 1:
The source information provision unit serves multiple functions: it tracks data sources for different language models, stores source information in a structured manner, and presents this information to users. This multi-functional component handles the complexity of multiple data sources centrally, allowing the system to provide reliability assessment without proportionally increasing overall system complexity.
Solution Approach 2:
The source information structure is nested within the response provision system, with source data organized in a hierarchical manner that links specific responses to their underlying data sources. This nested organization allows the system to manage multiple language models and their respective data sources efficiently, reducing the apparent complexity for users.
3Reliability
If source information is provided with each response, then users can understand the reliability of information, but the amount of information provided to users increases
Solution Approach 1:
The patent extracts source information as a separate, distinct element from the main response text. Instead of embedding source details within the response content, the system extracts and presents source information separately, allowing users to access reliability information without it increasing the volume of the primary response content.
Solution Approach 2:
The system adds source information in a different dimension - rather than increasing the length or complexity of the response text itself, source information is provided in a separate informational dimension. This allows users to access reliability data without the main response becoming more voluminous, effectively separating the quantity of response content from the quantity of source information.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An information providing apparatus according to an embodiment includes a generation unit and a providing unit. The generation unit generates a response to a prompt that has been input by a user by using one of a plurality of language models in each of which an amount of data sources that have been used for a training differs from one another. The providing unit provides the response generated by the generation unit to the user together with information that indicates a providing source of the data sources that have been used for the training in the language model and that has been used to generate the response.