LLM Control Agent with Segmented Manual Linking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The use of large language models (LLMs) in system control can lead to hallucination due to incorrect partial text references, resulting in inaccurate system control.
Innovation Solution
A computer system that supports control of a control target system by using a large language model, which includes an agent connected to the text generation system and a control tool for managing control manual information and link information. The agent executes processes to input control information, transmit inquiry text, execute control tools, and output answers, while ensuring accurate partial text references.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If the control manual is divided into partial text chunks with predetermined size, then the text amount that can be received by the LLM is limited, but incorrect partial text may be retrieved leading to hallucination
Solution Approach 1:
The control manual is divided into partial text chunks with predetermined sizes to fit within LLM input limits. The system manages these segments through control manual information and link information that tracks which segments are relevant to each control tool, enabling the LLM to process manageable portions while maintaining overall system control accuracy.
Solution Approach 2:
The system introduces control manual information and link information as intermediary structures between the control tools and the LLM. These intermediaries manage the relationships between control tools and partial text segments, ensuring that the LLM receives correct and relevant text references rather than incorrect segments, thereby preventing hallucination.
2Ease of manufacture
If the text is divided based on number of letters or paragraphs without considering context, then the text processing is simplified, but context loss occurs leading to incorrect partial text retrieval
Solution Approach 1:
The system applies different division strategies to different parts of the control manual based on their specific characteristics and context requirements. Rather than uniform division, the system considers local context quality when dividing text, ensuring that segments maintain sufficient contextual integrity for accurate control tool operation while still meeting LLM input size requirements.
3Adaptability or versatility
If the LLM processes large amounts of text, then the system control can be more comprehensive, but the text processing capacity of the LLM is limited
Solution Approach 1:
The control manual is segmented into manageable partial text chunks that fit within LLM input capacity limits. The link information structure enables the system to retrieve and process only the relevant segments needed for each control tool operation, allowing comprehensive system control coverage while respecting LLM processing capacity constraints.
Solution Approach 2:
The system retrieves and processes only the necessary partial text segments required for each control tool rather than loading the entire control manual. This partial action approach enables comprehensive system control capability while keeping individual LLM processing tasks within acceptable text quantity limits.
Data Source
AI summary
A computer system is connected to a text generation system including a control target system and a large language model and retains control manual information for managing a control manual of the control target system and link information for managing a relation between the control tool and partial text obtained by dividing the control manual. In the link information, second data associating the control tool with identification information of the partial text is stored. The computer system executes the control tool and transmits inquiry text including an execution result of the control tool to the text generation system when the response text including information regarding the control tool is received.