Language Model Instruction Execution With Executability Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural network models in terminal devices struggle to accurately recognize user commands, leading to inefficient and poor user experience due to repeated interactions required for correct command recognition.
Innovation Solution
An instruction execution method using a language model that performs multiple rounds of executability verification on candidate execution instructions until a valid instruction is obtained, ensuring high confidence and accuracy without additional user interaction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a neural network model is used to recognize user commands, then the terminal device can perform operations based on recognized commands, but the recognition accuracy is insufficient requiring multiple rounds of user rephrasing and interaction
Solution Approach 1:
The patent implements a feedback mechanism where the terminal device receives feedback information from the user about the execution result of the operation, uses this feedback to generate a new prompt, and feeds the new prompt back to the language model for improved command recognition. This iterative feedback loop continues until the command is accurately recognized and executed, resolving the contradiction between recognition accuracy and interaction efficiency.
Solution Approach 2:
The terminal device performs self-service by automatically generating improved prompts based on execution results and feeding them back to the language model without requiring manual intervention. The system self-corrects recognition errors through automated prompt generation and execution result analysis, improving accuracy while maintaining interaction efficiency.
2Reliability
If multiple rounds of interaction are required for correct command recognition, then the terminal device can improve recognition accuracy, but the user experience deteriorates due to repeated rephrasing requirements
Solution Approach 1:
The system implements an automated feedback loop where execution results are automatically used to generate improved prompts for the language model. This eliminates the need for manual rephrasing by users, as the system automatically iterates until accurate recognition is achieved, thereby improving user experience while maintaining high recognition accuracy.
Solution Approach 2:
The patent introduces an intermediary mechanism (the automated prompt generation system) that mediates between the user's initial command and the final accurate recognition. This intermediary automatically handles the iteration and refinement process, shielding the user from repeated rephrasing requirements while ensuring accurate command execution.
3Reliability
If the terminal device automatically iterates through multiple rounds of command recognition, then recognition accuracy improves, but the system complexity increases
Solution Approach 1:
The patent makes the terminal device multi-functional by enabling it to not only execute operations but also automatically generate improved prompts based on execution results and feed them back to the language model. This universal capability allows a single system to handle both operation execution and iterative prompt optimization, improving accuracy without proportionally increasing complexity.
Data Source
AI summary
An instruction execution method includes: receiving demand information of a target object; executing at least one round of target operation until a candidate execution instruction passing executability verification is obtained; in a first round of target operation, invoking the language model to perform instruction recognition on the demand information, and performing executability verification to determine whether the candidate execution instruction of the first round passes executability verification; in an ith round of target operation, invoking the language model to perform the instruction recognition according to a candidate execution instruction of an (i−1)th round of target operation and a guidance prompt information, followed by executability verification on the candidate execution instruction of the ith round of target operation; and using the candidate execution instruction passing executability verification as a target execution instruction, and controlling a terminal to execute an operation task corresponding to the demand information.