LLM UI Data Modification Using Typed Expressions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data entry and manipulation methods in software applications, such as ERP systems, rely on user input and are hindered by security risks, data communication restrictions, and substandard accuracy of large language models (LLMs) in performing mathematical operations.
Innovation Solution
A system that uses a Large Language Model (LLM) to generate a list of expressions for data manipulation in a user interface screen by sending only identifiers and data types, rather than actual data, and evaluates these expressions to perform the desired operations, enhancing security and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If actual data is sent to LLM for data manipulation, then data manipulation capability is improved, but security risks increase due to sensitive data exposure
Solution Approach 1:
The patent extracts only the essential elements (identifiers and data types) from the complete data set and sends them to the LLM. This extraction approach maintains the LLM's ability to manipulate data while removing sensitive information, thereby resolving the contradiction between manipulation capability and security.
Solution Approach 2:
The system introduces an intermediary layer that translates user requests into LLM prompts containing only identifiers and data types. This intermediary prevents direct exposure of sensitive data to the LLM while still enabling effective data manipulation through the generated expressions.
2Measurement precision
If complete data context is sent to LLM, then data manipulation accuracy is improved, but data communication restrictions are violated due to data size limits
Solution Approach 1:
The patent extracts only the critical components (identifiers and data types) needed for accurate data manipulation, excluding redundant or excessive data. This extraction maintains manipulation accuracy while adhering to data size restrictions.
Solution Approach 2:
The system applies partial action by sending only the necessary portion of data (identifiers and types) rather than the complete data context. This partial approach is sufficient for the LLM to generate accurate manipulation expressions without violating size constraints.
3Extent of automation
If LLM is used for mathematical operations, then automation is improved, but measurement precision deteriorates due to substandard accuracy
Solution Approach 1:
The system introduces an intermediary evaluation step where generated expressions are parsed and evaluated in a controlled environment. This intermediary layer verifies the accuracy of LLM-generated mathematical operations, maintaining high automation while ensuring precision through validation.
Solution Approach 2:
The system implements feedback through the evaluation and parsing of generated expressions. The evaluation process provides verification feedback on the correctness of LLM-generated mathematical operations, allowing for correction and improvement of accuracy while maintaining automation.
Data Source
AI summary
In an example embodiment, rather than send the state of the UI screen as a whole to an LLM to generate or modify one or more values on the UI screen, the identifiers and data types of data on the UI screen are gathered and sent to the LLM. The LLM is instructed to return a list of expressions that define what needs to be done to satisfy a user prompt. The calling process then evaluates the list of expressions to actually manipulate the data content.