Electronic Apparatus Prompt Selection for LLM Response Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Language models struggle to provide accurate and complete responses in real-time due to outdated training data and limited domain expertise, often requiring users to repeat inputs and resulting in lower user satisfaction.
Innovation Solution
An electronic apparatus that acquires user input, selects a sample prompt, and retrieves additional information from a server to generate a comprehensive input prompt, which is then used to provide responsive information to the user.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a language model is trained using past learning data, then the model can provide responses based on learned patterns, but the model cannot reflect real-time changing information and generates output based on outdated data
Solution Approach 1:
The system pre-generates multiple candidate responses using the language model before the user actually needs them. When user input is received, the pre-generated responses are immediately available for selection and refinement, eliminating the need for real-time generation delays and ensuring up-to-date information is provided without sacrificing response quality.
2Productivity
If the language model processes user input with limited information, then the processing speed is fast, but the model fails to provide complete and accurate response information
Solution Approach 1:
The system merges the language model's generated response with additional information retrieved from external data sources. The language model provides the structural framework and initial response, while external sources supplement with missing details, ensuring both speed and completeness are achieved through combined processing.
Solution Approach 2:
An intermediary information retrieval system is introduced between the user input and the language model processing. This intermediary quickly gathers necessary additional information and passes it to the language model, enabling the model to generate complete responses without being bottlenecked by information gathering delays.
3Reliability
If the language model requests the user to repeat utterance for insufficient information, then the model can clarify understanding, but the user feels uncomfortable and satisfaction decreases
Solution Approach 1:
The system proactively retrieves additional information and generates multiple candidate responses before the user even submits their input. By the time the user provides input, the system already has prepared responses that accurately reflect the intended meaning, eliminating the need for users to repeat themselves and maintaining interaction smoothness.
4Reliability
If the language model is trained in a specific field, then the model performs well in that domain, but the model fails to provide accurate responses in fields outside its training range
Solution Approach 1:
Domain-specific knowledge bases and external data sources are introduced as intermediaries between the general language model and user queries. When a query falls outside the model's training domain, the intermediary retrieves specialized information from appropriate sources, enabling accurate responses across multiple domains while maintaining the model's core strengths.
Solution Approach 2:
The system is designed to handle multiple domains by combining a universal language model with domain-specific information retrieval capabilities. The same core model serves all domains, but adapts to each domain by accessing specialized data sources, achieving both domain-specific accuracy and multi-domain versatility through a single unified system.
Data Source
AI summary
An electronic apparatus, including a memory configured to store a plurality of sample prompts; a communication interface configured to communicate with a server including a large language model; and at least one processor configured to: acquire user input of a user, select a sample prompt corresponding to the user input from among the stored plurality of sample prompts, acquire additional information corresponding to the user input, acquire an input prompt based on the user input, the sample prompt, and the additional information, transmit the input prompt to the server using the communication interface; and based on receiving response information corresponding to the input prompt from the server through the communication interface, provide the response information to the user.