LLM Personalization via User Correction Prompts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional large language models (LLMs) lack the ability to provide personalized responses to users and fail to learn from user interactions, leading to repeated mistakes.
Innovation Solution
A computer-implemented method that receives a task prompt from a user, identifies the context of the user input, determines a user correction prompt based on user changes to prior LLM outputs, and provides this conditioned prompt to the LLM to generate personalized responses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional LLMs are used without personalization, then the system complexity remains low, but the response accuracy and user satisfaction deteriorate due to inability to learn from user interactions
Solution Approach 1:
The system segments the LLM interaction into distinct components: the base LLM processing and the correction prompt mechanism. By separating these functions, the system achieves personalization without requiring complex modifications to the underlying LLM architecture, thus improving response accuracy while controlling system complexity
Solution Approach 2:
A correction prompt mechanism is introduced as an intermediary between the user input and the LLM. This intermediary captures user corrections and feeds them back to the LLM, enabling personalized responses without directly modifying the LLM's internal structure or requiring retraining
2Measurement precision
If LLM parameters are customized or retrained for personalization, then the response accuracy improves, but the computational resources and time required increase significantly
Solution Approach 1:
The system performs preliminary actions by collecting and storing user corrections in a correction prompt database before they are needed. When generating personalized responses, the system retrieves pre-processed correction information rather than performing real-time analysis or retraining, thus improving response accuracy while minimizing time loss
Solution Approach 2:
Instead of investing significant computational resources in permanent LLM parameter customization or retraining, the system uses lightweight, disposable correction prompts that are generated and discarded after each interaction. This approach achieves personalization at minimal computational cost and time
3Adaptability or versatility
If LLM parameters are customized for each user, then the personalization accuracy improves, but the device complexity and computational resources required increase
Solution Approach 1:
The correction prompt mechanism serves multiple functions: it captures user corrections, stores them in a database, retrieves relevant corrections, and feeds them back to the LLM. This universal mechanism enables personalization across different users and tasks without requiring separate customization processes for each user, thus improving adaptability while controlling device complexity
Solution Approach 2:
Instead of creating unique customized LLM parameters for each user, the system creates copies of correction information from the correction prompt database and feeds them to the LLM as context. This copying approach enables personalization for multiple users without requiring complex parameter customization, thus improving adaptability while maintaining simple system architecture
Data Source
AI summary
A method includes receiving a task prompt representative of a user input from a user and identifying, based on the task prompt, a context of the user input. The task prompt specifies a task for a large language model (LLM) to perform responsive to the user input. The method also includes determining, based on the context of the user input, a user correction prompt including one or more user changes made by the user to one or more prior outputs of the LLM. The method also includes providing, as input to the LLM, the task prompt conditioned on the user correction prompt to cause the LLM to generate a personalized response to the user input and providing the personalized response to the user input for output from a user device associated with the user.