Intermediate Agent Prompt Sharing Across Multiple LLMs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The high load imposed by generating prompts when communicating with multiple large language models (LLMs) due to the need to identify and generate prompts for each LLM independently, especially when large amounts of information are exchanged, is a challenge.
Innovation Solution
A chat control system that includes shared information generation and storage, allowing prompts to be generated and shared across LLMs, reducing the need for individual prompt generation by using an intermediate agent system to manage communication and store shared information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If information is acquired through communication between terminal and multiple LLMs independently, then various types of information can be acquired, but the load for generating prompts becomes very high
Solution Approach 1:
The patent merges the prompt generation function into a unified intermediate agent that handles communication with multiple LLMs. Instead of generating separate prompts at the terminal for each LLM, the intermediate agent consolidates information requirements and generates unified prompts, reducing the overall prompt generation load while maintaining the ability to query multiple LLMs for various information types.
Solution Approach 2:
The patent introduces an intermediate agent as a mediator between the terminal and multiple LLMs. This intermediary component manages the complex task of identifying information requirements, generating appropriate prompts for different LLMs, and consolidating responses, thereby reducing the burden on the terminal and enabling efficient multi-LLM interaction.
2Ease of operation
If information is acquired through communication to and from terminal and LLM, then interactive information acquisition is enabled, but identifying and generating prompts for second LLM requires additional processing load
Solution Approach 1:
The intermediate agent performs preliminary actions by maintaining a memory of previously acquired information and understanding the contextual relationships between different LLMs. When a user queries the second LLM, the agent can quickly identify required information from stored context rather than starting from scratch, significantly reducing the time needed for prompt generation while preserving interactive information acquisition.
3Adaptability or versatility
If multiple LLMs are used independently, then diverse information can be obtained, but the system complexity increases due to independent prompt generation for each LLM
Solution Approach 1:
The intermediate agent is designed as a universal component that handles communication with multiple different types of LLMs through a unified interface. It performs multiple functions including information requirement analysis, prompt generation for different LLM formats, response consolidation, and memory management, thereby reducing system complexity while maintaining the ability to obtain diverse information from multiple LLMs.
Data Source
AI summary
Provided is an intermediate agent system which stores shared information generated based on at least one of a chat message received from a user terminal or an answer received from a first LLM in a shared information storage module. The intermediate agent system outputs, to a second LLM, a prompt including an instruction to refer to the shared information storage module based on the chat message received from the user terminal. The intermediate agent system outputs, to the user terminal, a chat message based on an answer from the second LLM in response to the prompt.