Database Entry Generation Using Selective User Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The generation of database entries representing user inputs or information relevant to common user requests is time-consuming, especially when requestors are ambiguous, and large language models can be prone to errors and computationally expensive.
Innovation Solution
Breaking down the database entry generation process into multiple steps allows for requester feedback to be injected computationally inexpensively, using a natural language model to generate database entries organized as multiple elements, and updating these entries based on user feedback.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If large language models are used to generate database entries, then generation speed is improved, but computational cost increases and errors (hallucinations) occur
Solution Approach 1:
The patent segments the database entry generation process into multiple stages: initial entry generation using a smaller language model, followed by selective refinement stages where only specific elements are updated based on user feedback. This segmentation allows the system to use computationally cheaper models for the majority of generation work while reserving expensive large language model computations only when necessary for refinement, thereby reducing overall computational cost while maintaining productivity.
2Manufacturing precision
If multiple iterations of database entry development are performed, then accuracy is improved, but time consumption increases
Solution Approach 1:
The patent implements preliminary action by generating an initial database entry structure automatically using a language model before user feedback is incorporated. This preliminary generation provides a solid foundation that reduces the number of iterative cycles needed. Users can then refine specific elements rather than starting from scratch in each iteration, significantly reducing time consumption while maintaining accuracy through targeted improvements.
Solution Approach 2:
The system incorporates feedback mechanisms where users can provide input on specific elements of the database entry, and the system uses this feedback to generate refined versions of only the affected elements. This selective feedback approach allows for improved accuracy in specific areas without requiring complete re-generation of the entire entry, thereby reducing overall time consumption compared to traditional multi-iteration approaches.
3Manufacturing precision
If database entries are generated manually through communication between developer and requestor, then accuracy is improved, but time consumption increases
Solution Approach 1:
The patent implements self-service by enabling the system to automatically generate initial database entries using language models without requiring extensive manual communication between developer and requestor. The system can autonomously create the initial structure and even generate user interface components, significantly reducing the time spent on manual coordination while maintaining accuracy through automated intelligent generation rather than manual typing.
4Manufacturing precision
If user feedback is inserted into model-based generation process, then accuracy is improved, but process complexity increases
Solution Approach 1:
The patent segments the feedback integration process into discrete stages where user feedback is processed separately from the main generation process. The system identifies which elements of the database entry need refinement based on feedback, then generates updated versions of only those specific elements. This segmentation isolates the complexity of feedback processing from the overall generation process, making the system more manageable while improving accuracy through targeted refinements.
Data Source
AI summary
Embodiments are provided herein that include receiving, from a natural language model and based on a first textual prompt, a first output that represents a database entry, wherein the database entry includes a plurality of elements; obtaining a selection of a subset of the plurality of elements; determining a second textual prompt based on the selected subset of the plurality of elements; and generating, via the natural language model, a second output based on the second textual prompt, wherein the second output represents an update to the database entry. These embodiments provide for faster generation of catalog items or other types of database entries using generative natural language models in a manner that exhibits reduced memory requirements, computational cost, and/or amounts of training data.